Exchanges of innovation resources inside venture capital portfolios
Juanita González-Uribe*
February 2019
Abstract
I explore the prevalence of exchanges of innovation resources inside venture capital portfolios. I show
that after companies join investors’ portfolios, several proxies of exchanges between them and portfolio
companies (relative to matched nonportfolio companies) increase by an average of 60%. The increase
holds when joining events are plausibly exogenous and when VCs’ bargaining power and potential
conflicts of interest are low. Three novel mechanisms are supported: carve-outs, spawning, and
recycling, whereby entrepreneurs divest innovation units, start new ventures, and reuse assets in other
portfolio companies, respectively. Results suggest that returns to innovation are higher in venture
capital portfolios.
JEL classifications: G24, G32, O31, O34
Keywords: Venture capital portfolios, Innovation resources, State pension funds
* London School of Economics, e-mail:[email protected]. I thank Ulf Axelson, Bruce Kogut, Daniel
Ferreira, Camilo Garcia, William Greene, Dirk Jenter, Daniel Paravisini, Morten Sørensen, Scott Stern, Daniel
Wolfenzon, the editor William Schwert, and an anonymous referee for detailed comments. I also thank seminar
participants at the FED/NYU Stern Conference on the Role of Private Equity in the US Economy, the Searle Center
Conference on Innovation and Entrepreneurship, the Conference of Corporate Finance at the Olin Business School,
Washington University at St. Louis, Columbia University, Universidad de los Andes Colombia, Universidad
Católica de Chile, Duke University, University of North Carolina Kenan-Flager Business School, University of
Hawaii at Manoa, Harvard Business School, The Wharton School of the University of Pennsylvania, London
School of Economics, London Business School, University of California at Berkeley, University of Washington
Foster Business School, the Federal Reserve Board, and the Copenhagen Business School. Financial support from
the Ewing Marion Kauffman Foundation, the Coller Institute of Private Equity, and the Abraaj Group is gratefully
acknowledged. Xiang Yin provided excellent research assistance.
1
1. Introduction
Young companies’ difficulties in profiting from inventions have long been recognized by
academics and policy makers alike. In almost all cases, the successful commercialization of an
innovation requires the inventor’s knowhow to be combined with other innovation resources that the
original inventor lacks and seldom develops herself (Teece, 1994).1 The challenge lies in the many
frictions that can obstruct exchanges of innovation resources among young firms. For example, firms’
investments may be constrained by information asymmetries, expropriation risk can prevent inventors
from selling their inventions (cf. Arrow, 1975), and conflicts in arranging complete contracts can
preclude trade (cf. Allen and Phillips, 2000).
In this paper, I explore the prevalence of exchanges of innovation resources between companies
sharing common venture capitalists (VCs). As strategic investors, VCs have incentives to finance firms
with complementary innovation resources in an effort to increase investment returns; for example, by
internalizing innovation spillovers (cf. Teece, 1980; Hellman, 2002) or increasing product prices (cf.
Azar, Schmalz, and Tecu, 2018). In theory, VCs can also facilitate innovation exchanges; for example,
by punishing expropriation behavior (e.g., as advisors and board members; see Lerner, 1995 and
Hellman and Puri, 2002), bridging information asymmetries (e.g., in their role of screeners and
monitors; see Sørensen, 2007 and Kaplan and Stromberg, 2001, 2002), or financing otherwise
constrained firms. In practice, however, competition for investors could also obstruct collaboration
inside investors’ portfolios (cf. Fulghieri and Sevilir, 2009). Ultimately, how prevalent exchanges of
innovation resources are inside VC portfolios is an open question.
The main obstacle in exploring this question is that directly observing exchanges of innovation
resources, especially among young, private companies (the usual targets of VCs) is not possible: no
public markets for innovation resources exist. However, I can, and do, look for empirical evidence
inside venture capital portfolios of measurable cross-company interactions that typically lead to the
successful commercialization of industrial innovations (as validated by prior work), such as patent
citations (Hall, Jaffe, and Trajtenberg, 2005; Kogan et al., 2017), patent reassignments (Akcigit, Celik,
1 For instance, the commercialization of computer hardware typically requires specialized software.
2
and Greenwood, 2015; Serrano, 2010; Hochberg, Serrano, and Ziedonis, 2018), worker mobility
(Almeida and Kogut, 1999; Kaiser, Kongsted, and Rønde, 2015; Azoulay, Graff Zivin, and Sampat,
2012), strategic alliances (Mowery, Oxley, and Silverman, 1996; Stuart, 2000; Gomes-Casseres,
Hagedoorn, and Jaffe, 2006), and mergers and acquisitions (cf. Teece, 2010; Seru, 2014). I use an event
time framework that exploits differences in both the timing when and the investors from which
companies secure venture capital. I focus on innovative US companies that were financed by US VCs,
and filed at least one patent in the US Patent and Trademark Office during the 1976-2008 period.
The results show that after companies join the portfolio of a VC for the first time, several proxies
of exchanges of innovation resources between them and other companies in the VC’s portfolio
(portfolio exchanges) increase by an average of 60% (over the sample mean) relative to exchanges
between them and matched nonportfolio companies (nonportfolio exchanges). The increase holds for
both portfolio exchanges led by joiners and those led by portfolio firms, and is robust to excluding
companies with within-portfolio alliances (cf. Lindsey, 2008). The data support three novel mechanisms
of within-portfolio exchange: carve-outs—whereby restructuring companies divest some of their
innovation units inside the portfolio, spawning—whereby entrepreneurs move on to start new
companies also financed by the same VCs, and recycling—whereby the residual assets of restructured
companies are absorbed by other portfolio firms.
My interpretation of these results is that portfolio exchanges can be facilitated and used as a basis
for investment selection by VCs. Consistent with potential selection effects, there is evidence that some
portfolio exchanges begin to increase before the joiners enter the portfolio. On the other hand, the
relative increase in portfolio exchanges is concentrated in some situations where, absent common VCs,
such exchanges may not arise. For example, the increase is larger for exchanges financed by portfolio
companies that are in the same industries as joiners, and thus possibly subject to expropriation risk and
negative product rivalry effects, unless a common VC serves as arbiter.
Other more mechanical explanations are less consistent with the findings. For example, potential
industry, scale, technology, and location clustering effects have limited ability to explain the results.
Nonportfolio and portfolio companies are matched by technological focus, size, industry, and
3
geography, and any fixed differences between them are absorbed by the difference-in-differences nature
of the methodology. Unobserved differences between companies in and out of the venture capital
industry cannot explain the findings either: results continue to hold when I instead use a matching
methodology based on amounts of venture funds raised as well as company age, location, and
technology. A battery of robustness checks shows that results are also not driven by influential
observations at the state, firm, or company levels or by serial correlation and spurious trends.
The main implication of the results is that exchanges of innovation resources appear prevalent in
venture capital portfolios. As the successful commercialization of inventions typically requires such
exchanges, results suggest that returns are higher when innovations are backed by VCs.2 This suggestion
is consistent with the salience of VCs in the finance of high-value innovation: since the 1980s, patents
awarded to venture capital-backed companies amount to circa 6% of US patent grants and command
twice as many citations (a standard proxy of innovation value) as patents awarded to other types of
owners (see Online Appendix 1 and González-Uribe, 2013). However, whether such potential higher
returns trickle down to original inventors is unclear and hard to test: VCs face a complex set of
incentives, and no comprehensive data exist on return distribution between investors and founders.
While alternative interpretations cannot be fully ruled out, additional results provide some supporting
evidence that potential returns trickle down to founders. In particular, I show that exchanges continue
to increase in situations where VCs’ conflicts of interest and bargaining power over founders are
plausibly low (such as when the portfolio companies involved in the exchange with the joiner are mature
firms that the VCs have likely already exited because they entered the portfolio more than five years
prior).
The question remains: if companies were randomly assigned across venture capital portfolios,
would sharing a common VC causally facilitate exchanges of innovation resources? If so, this would
constitute evidence of an additional channel through which VCs could add value to their investments
(cf. Sørensen, 2007) and provide some support for the policies worldwide that encourage the
2 This suggestion is reminiscent of theories on how intangible assets can be more valuable when they coevolve
in a coordinated way with complementary assets, as seen in Rosenberg (1972).
4
development of venture capital markets (see Lerner, 2009). Answering this question is, however,
challenging given the low feasibility of a large scale randomized trial in the venture capital setting.
Here, I exploit the staggered adoption of “prudent man rules” (PIR), which allow local pension
funds to invest in venture capital across states in the US, as plausible exogenous variation in the
composition of local VCs’ portfolios (cf. González -Uribe, 2013).3 I estimate that a state’s PIR adoption
increases the capital commitments to the local venture capital industry that are made by local state
pension funds by 175 million USD (relative to pension funds located elsewhere), possibly because of
home bias in state pension funds’ venture capital investments (see Hochberg and Rauh, 2013).
I show that this influx in local venture capital commitments indeed changes the composition of
local VCs’ portfolios. PIR adoption in a state roughly doubles the mean probability that joiners first
enter local VCs’ portfolios, conceivably because these investors have more capital to invest, rather than
because of happenstance shocks to the joiners’ potential for cross-company exchanges. This influx also
increases (roughly triples) relative portfolio exchanges, even in subsamples of joiners and portfolio
companies that are located outside VCs’ home states (and further, also outside of “coincidental states”
that adopt PIR at the same time as the home states of VCs), which mitigates concerns that results are
driven by the endogeneity of PIR adoption (or the potential impact of PIR adoption on other local
investments not related to venture capital). I present evidence against other methodological concerns
such as aggregate trends in PIR adoption, California and/or Massachusetts effects, and biases from mis-
measurement in the actual implementation dates of PIR adoption across states. Under the assumption
that PIR adoption in VCs’ home states does not disproportionately increase relative portfolio exchanges
through channels other than portfolio joining events, these additional results constitute evidence that
sharing common VCs causally facilitates exchanges of innovation resources.
This paper contributes to the literature on organizational design and innovation (e.g., Seru, 2014;
Branstetter and Sakakibara, 2002; Spence, 1984; Bernstein and Nadiri, 1989). As described by Teece
3State pension funds are regulated at the state level and were thus not covered by the Employment Retirement
Income Security Act of 1979 that clarified how pension funds ran at the federal level could invest in venture
capital (see Kortum and Lerner, 2000).
5
(2010), the most significant transformation in the organization of corporate innovation during the last
five decades has been the shift away from centralized laboratories (popular in the 1950s) toward new,
more decentralized models such as horizontal and vertical alliances across firms. My contribution is to
show that VCs appear to select their investment portfolios based on the potential for these organizational
models. Moreover, the PIR adoption results suggest that VCs can also facilitate some of these
organizational models among young companies, where, absent common owners, frictions such as
information asymmetries may prevent them from arising on their own.
This paper also contributes to the literature on the role of venture capital in the real economy (e.g.,
Kortum and Lerner, 2000; Mollica and Zingales, 2007; Hirukawa and Ueda, 2008; Popov and
Rosenboom, 2009; Bernstein et al., 2016). Most related work quantifies the contribution of VCs but
remains agnostic about how VCs contribute. My results point to one potential mechanism through
which venture capital can influence innovation: appropriation of innovation returns inside VCs’
portfolios. This mechanism is consistent with theories on how investors’ portfolios provide
complementary resources to support venture capital-funded firms (Hellman, 2002) and are reminiscent
of prominent VCs’ claims on how common cross-company collaborations are in their portfolios.
Relative to the wider literature on private equity, results show that the role of this industry in the
redeployment of innovative assets to efficient use extends to venture capital and is not confined to the
long-documented role of buyout funds (cf., Jensen 1989; Kaplan and Stromberg, 2009).
My work is most closely related to Lindsey (2008), who shows that strategic alliances are
disproportionately likely among companies sharing common VCs. Using a different methodology and
sample, I show that the potential role of VCs in expanding firm boundaries is much more extensive than
previously known and how it ranges from informal knowledge exchanges to more formal transfers of
human and other capital resources. In addition, I offer new evidence on the timing of these exchanges,
make a first pass at parsing selection from causal effects, and introduce three novel complementary
mechanisms of within-portfolio exchange. My work also relates to Gompers and Xuan (2012), who
6
explore mergers and acquisitions when the bidder and target share a common VC. There, however, the
focus is on merger announcement returns and the structure of such transactions.4
The rest of this paper proceeds as follows. In Section 2, I describe the data. In Section 3, I explain
how I measure portfolio exchanges. Section 4 describes the empirical strategy and shows that there are
significant increases in relative portfolio exchanges after joiners first enter VCs’ portfolios. In Section
5, I describe the PIR adoption effect on the local venture capital industry and on relative portfolio
exchanges. Section 6 concludes the paper.
2. Data
2.1 Capturing data on investments by VCs
My starting point is the universe of transactions that closed between January 1976 and December
2008 and are registered in Securities Data Company’s (SDC’s) VentureXpert database. Before 1976,
there are only a handful of publicly recorded investments by venture capital firms. While information
on more recent investments is available from the same source, I consider only investments made in or
before 2008, as many of my outcome variables can only be accurately measured up to that year (see
Section 3). In addition, while similar information is provided by Venture Source, a unit of Dow Jones,
I chose to use the SDC data, as Maats et al. (2011) and Kaplan and Lerner (2017) argue that the latter
source has better coverage of investments.
I eliminate three types of investments from the data: transactions by private equity groups other
than independent VCs (e.g., angel groups), transactions by venture capital firms that are not early stage
investments (e.g., buyout funds), and investments by VCs in companies that were already traded in
public markets before the transaction and secondary purchases. Finally, I only include investments made
by US VCs in US companies.
4 The paper also relates to Gonzalez-Uribe (2013), which shows that citations of patents increase after companies
secure venture capital. I focus here only on portfolio citations, which are less than 1% of citations and explain less
than 2% of the overall increase after companies secure venture capital (Gonzalez-Uribe, 2013).
7
For each VC and company pair, I keep track of the name of the fund through which the company
first joined the VC’s portfolio. In addition, I keep track of every VC’s composition of its investment
portfolio (i.e., all the companies that the investor has financed through its multiple funds). A company
is part of a VC’s investment portfolio from the first year it secures investment from such an investor
until the end of the sample.
The data contain 42,670 venture capital portfolio joining events, in which 19,156 joiners enter the
portfolio of one of 1,876 VCs for the first time between 1976 and 2008. All additional financing rounds
between a company and an existing venture capital investor are not counted as portfolio joining events
in the analysis.
2.2 Capturing data on innovating companies
I match the companies that secured venture capital to their patent records at the U.S. Patent and
Trademark Office (USTPO) based on company name. To do so, I use the Harvard Business School
(HBS) patent database (see Lai, D’Amour and Fleming, 2009) and the data by Kogan et al. (2017). The
HBS data contain all electronic records of the USPTO for patent filings between 1976 until 2010, which
have been cleaned and consolidated by HBS, including information on citations made and received by
these patents, as well as the names and locations of the inventors and of the assignees. I use the data by
Kogan et al. (2017) to complement information about assignee/inventor location whenever it is missing.
I restrict my sample to primary assignments of utility patents (99%) filed by U.S. companies through
December 2008, because I cannot construct citation data for more recent patents given application-to-
award lags at the USTPO (see Hall, Jaffe and Trajtenberg, 2001; Seru, 2014; Kogan et al., 2017). After
the restrictions, the sample consists of 2,881,097 patents awarded to 1,980,696 inventors and issued to
242,767 U.S. assignees.
2.3 Combining patent and venture capital information
To combine the two databases, I strip punctuation, capitalization, and common acronyms from
company names taken from VentureXpert and assignee names taken from the HBS database. I then
combine the samples on the normalized company and assignee names using a fuzzy match procedure
8
that scores potential matches based on the Levenshtein edit distance.5 Using a random sampling
procedure, I determine a score threshold such that matches with scores above the threshold are hand
checked and those below the threshold are eliminated. During the manual check of the remaining
matches, and whenever possible, I verify that the two companies are in the same state. In some
ambiguous situations, the names are similar but not identical, or the location of the patentee differs from
that given in the records of SDC. In these cases, I research the potential matches using web searches. In
some cases, multiple names in either of the databases appear to match a single name in the other data
set. For these, I add the observations into an aggregated entity.
After the name match, I manually go over observations that present inconsistencies in the
information retrieved from the SDC and patent data sets. For example, I exclude all patents that,
according to the patent data, were filed before the companies were founded (as recorded by the SDC).6
Patents filed by Sun Microsystems make up 9.0% of the patents filed by venture capital-backed
companies. In robustness checks, I show that the results are the same if I exclude Sun Microsystems
from the analysis (see Section 3.2). After these exclusions, the patent awards of no single company
make up more than 5% of the data.7 Similarly, patents filed by companies funded by the venture capital
firm Kleiner Perkins Caufield & Byers constitute 15% of the data. In robustness checks, I show that the
results are the same if I also exclude these companies from the analysis (see Section 3.2). Financing
events by no venture capital firm or to no company constitute more than 5% of the sample [New
Enterprise Associates Inc. and Applied Micros Circuits Corporation are, respectively, the venture capital
firm and company with the next largest number of financing events, 247 (2%) and 14 (0.12%)].
5 The Levenshtein edit distance is a measure of the degree of proximity between two strings and corresponds to the number
of substitutions, deletions, or insertions needed to transform one string into the other one (and vice versa). 6 Further, I also exclude companies founded before 1976 and companies that filed innovations or were founded
more than ten years before they secured their first VC investment. For example, I exclude all patents assigned to
SmithKline Beckman Corporation, which was founded in 1830. Although investments in this company pass all
the restriction filters imposed before the match, this company does not correspond to the typical VC investment.
Not only had the company been established for more than 100 years before securing VC, but the investment was
by Faneuil Hall Associates, a family investment firm that was classified by SDC as a venture capital firm. 7 Other important companies in the sample in terms of the number of patents include 3Com Corporation (1.3%),
Altera Corporation (1.5%), Applied Materials Inc. (4.1%), Compaq Computer Corporation (1.8%), Cypress
Semiconductor (1.5%), Genentech Inc. (1.5%), SciMed Life Systems (1.6%), VLSI Technology Inc. (1.0%), and
Xilinx Inc. (1.8%).
9
2.4 Analysis sample
The final sample includes information for 3,960 joiners that filed 74,771 patents (approximately
2.5% of all patent awards) and first entered the portfolio of one of 1,265 VCs during 1976-2008. The
data include 11,815 portfolio joining events (approximately 28% of joining events in the SDC data).
The small size of the matched sample relative to the universe of patenting assignees is consistent with
the relatively small size of the venture capital industry. The modest sample size, relative to the universe
of joining events, is consistent with the prevalence of venture capital investments in industries where
intellectual property is not necessarily protected using patents (e.g., internet, media, and software
companies) and with the large attrition of companies in the SDC files (i.e., one-third of the companies
that secure VC financing are liquidated, many of them before the company files a patent).
I structure the data in “portfolio joining event time” for every joiner and VC pair in the sample. I
collect information on exchanges across the joiner and the portfolio companies of the VC five years
before and six years after the joining event. For some of the joining events early (late) in the sample I
can only collect information for a window of less than five years before (six years after) the investment.
In robustness checks, I show that the results are quantitatively similar if I restrict the data to joining
events for which information for the full 11-year window is available (see Section 4.2).
Table 1 presents the composition of the sample and summary statistics. The first five columns in
Panel A show the distribution of portfolio joining events over time by type of investment: total, seed,
early stage, expansion, and later stage. The last two columns in Panel A show the distribution over time
of patent applications and grants.
Panel A in Table 1 shows that the distribution of patent applications and portfolio joining events
mimics the increase and decline of the venture capital industry throughout the boom and bust of the dot-
com crisis, particularly early stage joining events. The distribution of grants reflects the lags in patent
awards by the USPTO.
Panel B in Table 1 shows the distribution of the sample across companies’ and VCs’ home states.
Not surprisingly, the sample is concentrated in California (56% of patents, 47% of companies, and 33%
10
of VCs), where most VC activity takes place in the US. In robustness checks I show that the results are
not driven by potential California effects (see Sections 4.2 and 5.4). Panel C shows the distribution of
patents in the sample across two-digit technology classes. The most popular technology classes include
Computer Hardware and Software (17%), Communications (11%), Semiconductor Devices (10%), and
Medical Instruments (9%). The importance of these technology classes in the sample reflects the
industry distribution of VC investments described in Panel D, which is concentrated in medical health
(18%), semiconductors (17%), computer software (16%), and communications (11%).
3. Measuring cross-company exchanges of innovation resources
Because exchanges of innovation resources between companies are not observable, I construct
several proxies based on company interactions related to the distribution and promotion of innovations.
Several such proxies have been associated by prior work to the successful commercialization of
inventions, such as patent citations, patent reassignments, worker exchanges, strategic alliances, and
mergers and acquisitions.
Patent citations are the standard metric in the innovation literature to measure knowledge exchanges
(Jaffe, Trajtenberg, and Henderson, 1993; Hall, Jaffe, and Trajtenberg, 2005; Kogan et al., 2017). These
citations serve the important legal function of limiting the innovation protected by the patent document.8
While patent citations are useful in that they allow for a paper trail between innovations, the downside
is that parties other than the inventors may add citations, and hence some citations may not necessarily
reflect actual knowledge flows between the cited and citing parties (Thompson and Fox-Kean, 2005;
Roach and Cohen, 2010). For example, it is possible that VCs encourage citations across companies in
their investment portfolio for legal reasons, such as building patent thickets that may restrict the access
of other unrelated parties to their knowledge. These strategic citations may also be consistent with cross-
company exchanges that can increase profits from inventions. The difference is that they would not
correspond to knowledge flows but rather to exchanges of intellectual property protection. Using
8 Arguably, this legal function creates strong incentives for inventors to get these citations right. As Jaffe,
Trajtenberg, and Henderson (1993) put it, including extraneous citations is “leaving money on the table.” Likewise,
deliberately excluding citations can expose the company to costly sanctions by the regulator or infringement
lawsuits (Branstetter, 2006).
11
information from the HBS patent data set and from Kogan et al. (2017), I construct three proxies of
exchanges of innovation resources based on patent citation data: Citations received, Citations made, and
Overall citations. These proxies measure over time, respectively, citations from the portfolio companies
to the patents of the new joiner, citations from the joiner to the patents of other portfolio companies, and
the combination of both of these types of citations. Given the lag between application grants and
applications, I only consider patents filed by 2008, so as to have at least two years of citation data (the
last year of filings in the patent data is 2010; see Section 2.2).
Patent reassignments are transfers of intellectual property between patent assignees and third
parties. They have been used in several papers as measures of ownership transfer of innovations between
companies (Akcigit, Celik, and Greenwood, 2016; Galasso, Schankermann, and Serrano, 2013;
Hochberg, Serrano, and Ziedonis, 2018; Serrano, 2010). I focus here on reassignments to other
innovative companies given the investment profile of venture capital firms. I note however, that other
important patent reassignments include collateral pledges to financial institutions (see Mann, 2018).
Based on data from the USTPO Bulk Downloads at Google, I construct three proxies of exchanges of
innovation resources using patent reassignments: Patents sold, Patents bought, and Patent sales.9 The
first (second) proxy measures over time the patents bought by (sold to) joiners from (by) other portfolio
companies. The third proxy combines the first two measures of patent reassignments.
Cross-company worker mobility mediates exchanges of innovation resources, such as knowledge
flows, between firms (Almeida and Kogut, 1999; Agrawal, Cockburn, and McHale, 2006; Azoulay,
Zivin, and Sampat, 2012; Kaiser, Kongsted, and Rønde, 2015; Fallick, Fleischmann, and Rebitzer,
2006). This type of exchange can be especially relevant in the venture capital context, as VCs are also
known to help companies hire (cf. Hellman and Puri, 2002). Worker mobility between two companies
is inferred here mostly from the inventor files of the patent data because comprehensive information on
the workers of venture capital-backed companies is not publicly available.10 I define an inventor as
9 See: https://www.google.com/googlebooks/uspto-patents-assignments.html 10 In complementary analysis, I also use information on company executives available at SDC; see Section 4.4
for more details.
12
moving from company A to company B at time t, if at time t the inventor assigns a patent to company
B, and at any year prior to t the inventor assigned a patent to company A. I construct three proxies of
exchanges of innovation resources using the HBS inventor data: Inventor emigrates, Inventor
immigrants, and Inventor exchanges. The first (second) proxy measures over time the number of
inventors that moved from the joiner (any portfolio company) to any of the portfolio companies (the
new joiner). The third proxy combines the first two measures. I note that these proxies are not meant to
capture, and do not necessarily capture, violations of noncompete agreements, which are common in
employment contracts (Marx, 2013; see also discussions in Sections 5.3 and 5.4). Rather, these measures
are more likely to measure the legal turnover of inventors across firms.
I also consider several types of business integration between joiners and portfolio companies,
including alliances as well as mergers and acquisitions. Alliances have been used as measures of
exchanges of innovation resources between firms (Mowery, Oxley, and Silverman, 1996; Stuart, 2000;
Gomes-Casseres, Hagedoorn, and Jaffe, 2006), particularly venture capital-backed firms (Lindsey,
2008), while mergers and acquisitions are a standard measure of firm integration in the finance literature
(Seru, 2014). I construct two final proxies of exchanges of innovation resources based on these
additional data: alliances and mergers and acquisitions. Each proxy equals one after the joiner and at
least one of the portfolio companies enter an alliance or merge (get acquired or are acquired),
respectively. I manually match the companies that secured venture capital to their alliances and merger
and acquisition activity based on company name. To do so, I employ information from SDC Platinum
at Thomson-Reuters.11
Table 2 shows summary statistics of the different proxies of exchanges of innovation resources
between joiners and portfolio companies in the sample. I refer to these exchanges as portfolio exchanges
throughout.
11 The data on alliances include 128,770 alliances between US-based companies for the 1972-2016 period. There
are a total of 17,366 alliances between innovating companies and 5,485 (761) in which at least one (both) of the
innovating companies is VC-backed. The data on mergers and acquisitions include 324,762 mergers and
acquisitions between US-based companies for the 1972-2016 period. There are a total of 95,341 mergers and
acquisitions between innovating companies and 38,563 (6,891) in which at least one (both) of the innovating
companies is VC-backed.
13
3.1 Patterns in portfolio exchanges between joiners and other portfolio companies
Fig. 1 plots the average portfolio exchanges between joiners and portfolio companies for the different
proxies against joining event time. The solid (dotted) line plots the estimated coefficients (95th
confidence intervals) of the 𝛽𝑘𝑠 in the following equation:
𝑦𝑖𝑗𝑡 = 𝛼𝑖𝑗 + ∑ 𝛽𝑘𝐸𝑣𝑒𝑛𝑡𝑖𝑗𝑡+𝑘
5
𝑘=−5
+ 𝛾𝑡 + 휀𝑖𝑗𝑡 , (1)
where 𝑦𝑖𝑗𝑡 is any measure of portfolio exchanges between joiner i and portfolio companies of VC j at
time t, and the 𝐸𝑣𝑒𝑛𝑡𝑖𝑗𝑡+𝑘𝑠 are event-time dummies that light up k years before/after the joiner first
joins the VC’s portfolio. Fixed effects for every pair of joiner and VC, 𝛼𝑖𝑗, absorb any time-invariant
complementarities between joiners and other portfolio companies. Calendar year fixed effects, 𝛾𝑡,
control for aggregate trends in the different measures of portfolio exchange.
The series of coefficients 𝛽𝑘 are the main estimates of interest. For 𝑘 > 0, 𝛽𝑘 is an estimate of the
change in portfolio exchanges k years after the joining event between portfolio companies and joiners,
relative to all other portfolio exchanges for joiners in the sample that did not enter the portfolio of a VC
k periods ago but did so before or will do so after. Similarly, for 𝑘 < 0, 𝛽𝑘 is an estimate of the change
in portfolio exchanges 𝑘 years before the financing event. The 𝛽𝑘 coefficients are normalized relative
to the year before the joining event (𝑘 = −1), which is set to zero. Standard errors are clustered at the
joiner and VC pair level to adjust for heteroskedasticity and within-pair correlation over time.
Fig. 1 shows that after joiners enter VCs’ portfolios for the first time, portfolio exchanges between
them and other companies in the portfolio increase. For example, Panel A shows an increase in Citations
received after the joiner enters the portfolio, while no trend exists prior to the joining event. The rest of
the panels in the figure plot portfolio exchanges between joiners and portfolio companies against joining
event time for the different proxies and show a similar increasing pattern.
The patterns in portfolio exchanges exhibited in Fig. 1 are consistent with the prevalence of
exchanges of innovation resources inside venture capital portfolios. Other potential explanations
14
naturally exist. For example, the patterns can also reflect trends in the size of venture capital portfolios
and joiners. Given that VCs often specialize their investments, the patterns may also reflect potential
industry, scale, technology, and location clustering effects.
To control for potential aggregate trends and clustering effects, I adjust all metrics of portfolio
exchanges using data on the different proxies of exchanges of innovation resources between joiners and
matched portfolio companies. For every company in the portfolio of a VC, I select an observationally
equivalent match from the universe of patenting companies in the US that did not secure venture capital
over the sample period. In particular, portfolio companies and their respective matches are located in
the same state and have the same patent-production scale (i.e., number of patents), technological base
(i.e., distribution across two-digit technology classes of citations made to prior innovations), and
technological focus (i.e., the three-digit technology class mode of patents).12 Because venture capital-
funded companies are likely to have different growth paths than non-venture-backed companies, the
matches are recalibrated every year.13 Online Appendix 2 has a detailed explanation of the matching
procedure and discusses matching statistics. Using these matched portfolio companies, I then estimate
relative portfolio exchange measures for all the outcome variables included in the analysis. For each
original measure of portfolio exchanges, the relative version subtracts (from the original) the respective
exchanges between joiners and matched nonportfolio companies. To distinguish between the different
types of exchange measures, I identify the matched measures with the prefix “matched,” and the relative
measures with the prefix “relative.” Table 2 reports summary statistics for the relative and matched
outcome variables.
Fig. 2 shows that potential aggregate trends or clustering effects cannot explain the increase in
portfolio exchanges after the portfolio joining event. Following the same structure as Fig. 1, Fig. 2 plots
relative portfolio exchanges between joiners and portfolio companies against joining event time for the
12 In Online Appendix 4, I show that results are quantitatively similar when I use alternative matching
methodologies based on the amount of venture funds raised, company, age, location, and technology (see also
Section 4.2). 13 In unreported results, I check whether the results are the qualitatively the same if I rely instead on a single match
across the sample for each portfolio company. I present the results using the annual matches because the matching
statistics are significantly better, as discussed in more detail in Online Appendix 2.
15
different proxies. The increasing pattern post-joining is still evident for the relative measures. For most
of the relative exchange metrics, there is a clear mean shift after joiners enter the portfolio (see Panels
A-I). For Relative alliances and Relative mergers and acquisitions, there is a sharp trend break after
joiners enter the portfolio (see Panels J and K, respectively). The patterns suggest that more
contractually difficult arrangements across joiners and portfolio companies (such as integration) are
only established with time, after the joiners settle in the portfolio.
Fig. 2 also shows that the increase in relative portfolio exchanges holds for both exchanges financed
by joiners and those financed by the other companies in the VC’s portfolio. After joiners first enter a
VC’s portfolio, portfolio companies are relatively more likely to cite (Panel A) and purchase (Panel D)
joiners’ patents as well as hire joiners’ inventors (Panel G). These results suggest that financing effects
where the increase in relative portfolio exchanges is due to the cash infusions of joiners can only
partially explain the results, as portfolio companies do not generally secure financing at the same time
as the new joiner. Other potential explanations include the mitigation of contracting complexities, which
are also prevalent among young companies, inside VC portfolios.
Fig. 2 also shows evidence of anticipation effects (that is, evidence of increasing relative portfolio
exchanges before the joiner first enters the portfolio), such as Relative citations made (Panel B) and
Relative inventor immigrants (Panel H). These patterns suggest that portfolio exchanges can be used as
a basis for selecting investment portfolios by VCs.14 For some of the metrics, the trend plateaus or
reverts after a couple of years.15 The timing of this plateau is likely explained by the exit of the VC from
the company (cf. Lindsey, 2008).
4. Empirical strategy
14 Unreported results suggest that VCs also use information on portfolio exchanges to select possible syndication
partners. I find that the prefunding patterns in Figs. 1 and 2 are more pronounced for the later rounds when the
investment syndicate increases than for the earlier rounds. This additional result is broadly consistent with the
work of Hochberg, Lindsey, and Westerfield (2015) on the importance of resource accumulation in the formation
of venture capital co-investment networks. 15 For example, the plots for Relative inventor emigrates (Panel G) and Relative inventor immigrants (Panel H),
respectively
16
I summarize the results of the portfolio joining event study analysis of Section 3.1 in Table 3. I report
results from a more parsimonious model than Eq. (1):
𝑦𝑖𝑗𝑡 = 𝛼𝑖𝑗 + 𝛽𝑃𝑜𝑠𝑡𝑖𝑗𝑡 + 𝛿𝑃𝑜𝑠𝑡𝑖𝑗𝑡 × 𝜏 + 𝛾𝑡 + 휀𝑖𝑗𝑡 , (2)
where 𝑃𝑜𝑠𝑡𝑖𝑗𝑡 is a variable that equals one after joiner i enters the portfolio of VC j for the first time.
The term 𝑃𝑜𝑠𝑡𝑖𝑗𝑡 × 𝜏 is the interaction between 𝑃𝑜𝑠𝑡𝑖𝑗𝑡 and the number of years since the time of the
joining event (𝜏). This interaction term will be equal to zero in the years before the joiner enters the
portfolio and start at one in the year after the joining event.
The specification in Eq. (2) allows for both a mean shift (𝛽) and a trend break (𝛿) after the joining
event, which captures the dynamics of joining venture capital portfolios more flexibly. The coefficient
of interest, 𝛽 (𝛿), measures the change in the mean (slope) outcome of interest before and after the
portfolio joining event. These changes are relative to all other joiners that do not join a new portfolio in
that year (but have either already entered a new portfolio or will join a new portfolio in the future).
I do not control for any time-varying variables at the level of the new joiner, portfolio, or pair, as
most of these are outcomes of the joining event. Therefore, I cannot separate the effect of entering the
portfolio from the effect of other changes that occur at the same time as the joining event but are not
related to the event. In all regressions, standard errors are clustered at the joiner and VC pair level to
adjust for heteroskedasticity and within-pair correlation over time. The results are quantitatively similar
with other levels of clustering (see Section 4.2).
4.1 Main results
Table 3 confirms the patterns in Figs. 1 and 2. After a joiner first enters the portfolio of a VC, there
is a positive and significant mean shift in portfolio exchanges (Panel A) as well as in relative portfolio
exchanges (Panels B) between the joiner and other companies in the VC’s portfolio. On average, the
different types of portfolio exchanges increase by 60% (over the sample mean) relative to nonportfolio
exchanges.16 The largest mean shift is for Relative inventor exchanges, which increase by 0.012
16 For example, the estimated number of additional portfolio citations received by new joiners within five years
of the joining event is 0.47 (0.02+0.03×1+…+0.03×5; see Panel A in Table 3).
17
(Column 9, Panel B), corresponding to a 120% increase over the sample mean (0.01; see Table 2). The
smallest significant mean shift is for Relative patents sold (Column 4, Panel B), which increase by
0.004, representing a 40% growth over the unconditional mean (0.01; Table 2). Online Appendix 3
shows that the increase in relative portfolio exchanges is similar on both the coasts, where VC funding
is common, as well as in the rest of the US.
The last two columns in Panel B of Table 3 show a trend break of 0.001 in both Relative alliances,
and Relative mergers and acquisitions, even though there is no mean shift. This trend break is
economically significant; it respectively corresponds to a 100% and 50% increase per post-event year
over the sample mean (0.001 and 0.002; Table 2). Evidence in support of a trend break but not in support
of a mean shift in Relative alliances simultaneously confirms Lindsey’s (2008) “alliance mechanism”
(using a different methodology and sample) and suggests that this mechanism of resource exchange is
not the only one at work within venture capital portfolios (as there is evidence of a mean shift for all
other measures except mergers and acquisitions). The results in Online Appendix 3 confirm this
suggestion by showing that the increase in portfolio exchanges holds after I exclude all joiners with
within-portfolio alliances from the analysis. I return to a discussion of potential mechanisms in Section
4.4.
4.2 Robustness checks
Results from several robustness checks are summarized in the Online Appendix. Online Appendix
3 shows the findings are not driven by state, firm, or company-level influential observations; most
results (20 out of 22) hold after I exclude observations from the states of California and/or Massachusetts
and from the firms Kleiner Perkins and Sun Microsystems. The results are also robust to excluding all
joiner and VC pairs for which I cannot construct the full financing event 11-year window (see Section
2.4).
Online Appendix 3 also shows that differential trends across industries or states do not drive the
results, which continue to hold after I include fixed effects at the level of industry by year, joiner state
by year, and VC state by year. They also hold for different levels of clustering at the VC level, joiner
18
level, or at both the VC and joiner levels. Concerns of serially correlated outcomes are also mitigated
in the Online Appendix, where I show that results are quantitatively similar after I collapse the time
series variation to a two-period pair panel (and include fixed effects for every joiner and VC pair in the
estimation). Because joining events occur at different years across joiner and VC pairs, I collapse the
data by using pairwise residuals from regressing 𝑦𝑖𝑗𝑡 (i.e., the proxy of exchanges of innovation
resources) on year dummies and aggregating those from years before and after the joining event into
the pre- and post- periods for every joiner and VC pair in the data (see Bertrand, Duflo, and
Mullainathan, 2004).
In Online Appendix 3, I also show that the results are not driven by a mechanical increase in VC
portfolio size over time nor by spurious time trends. I summarize results from 1,000 placebo tests where
I randomly pick the years of joining events for the pairs in the sample. Nonrejection rates are close to
(and often below) 5% for all measures of portfolio exchanges, as would be expected from randomly
choosing the joining events.
Finally, Online Appendix 4 shows that the results are not driven by differences between companies
in and out of the venture capital industry. The findings are quantitatively similar under alternative
matching methodologies based on the amounts of venture capital raised as well as on company age,
location, and technology.
4.3 Heterogeneity
Tables 4 and 5 report the regression results from several cuts of the sample. Consistent with VCs
potentially facilitating portfolio exchanges, Panels A and B in Table 4 show that the relative increase in
portfolio exchanges appears strongest (for most proxies) for joiners and pairs where the expropriation
risks and information asymmetries are largest. Most notably, the increase is stronger for joining
companies that are more likely to have uncertain products and technologies, such as younger firms
(Panel A) or those in an early stage of development (Panel B). Younger firms are those that are below
the median company age of two years at the time of financing. Company age is measured relative to
19
founding age. Early stage firms are those whose first investment by a VC was secured either at seed or
early stage according to the SDC files.
Also consistent with VCs facilitating exchanges of innovation resources, Panel A in Table 5 shows
that the increase in portfolio exchanges holds in situations where cross-company exchanges are likely
fraught with contracting complexities. Most notably, and consistent with Lindsey (2008), the increase
holds between industry competitors (i.e., joiners and portfolio companies that have the same SDC
industry classification).17
Panel D in Table 4 shows that the increase in relative portfolio exchanges is, on average, stronger
and materializes faster (i.e., there is a significant mean shift after joiners enter the portfolio) for more
experienced VCs, as measured by the number of prior investments made by the VC, following Lindsey
(2008). For younger VCs, the effect is instead more gradual; columns 1, 3, 4, 5, 6, 10, and 11 show a
positive and significant trend break (but no evidence of a mean shift) after joining young VCs’
portfolios. In unreported regressions, I confirm that the differences across senior and junior VCs are not
driven by differences in portfolio size across investors of different ages. These differences also hold for
“first rounds” where companies secure venture capital (from any investor) for the first time and appear
stronger for later (i.e., not the first) investment rounds. A VC is defined as young if its age is below the
median investor age of six years at the time of the investment, as measured relative to the first time in
the SDC data that the VC makes an investment.
Further, consistent with how competition for VCs’ resources may also prevent cross-company
exchanges, Panel B in Table 5 shows the increase in relative exchanges is often most pronounced
between joiners and other companies in the portfolio of the VC that are not in direct competition for the
VC’s resources because they were financed by a different fund. For joiners and portfolio companies
financed by the same fund of the VC, only relative portfolio exchanges related to patent reassignments
(columns 4-6, Panel B) are more likely to increase after the portfolio joining event.
17 In unreported exercises, I show that the results are similar across new joiners of “generalist” and “specialist”
VCs. I code a VC with a minimum of five investments as specialist (generalist) if 90% of its investments over the
previous five years are (not) focused on one SDC sector, following Hochberg, Mazzeo, and McDevitt (2015).
20
The main implication of results is that exchanges of innovation resources appear prevalent in
venture capital portfolios. Since the successful commercialization of inventions typically requires firms
to combine their inventions with other complementary resources that they do not own or develop, my
preferred interpretation of the results is that returns to innovation are higher when inventions are
developed inside venture capital portfolios. This interpretation is supported by prior literature showing
positive links between cross-company exchanges and value, such as studies showing how resource
exchanges across firms are met, on average, with positive reactions in the public equity markets (e.g.,
Hall, Jaffe, and Trajtenberg, 2005, for patent citations; Denis and Denis, 1995, for CEO turnover; Chan
et al., 1997, for alliances; and Kaplan, 2006 for mergers and acquisitions). However, I cannot fully rule
out that in the venture capital context, higher innovation rents do not trickle down to inventors: VCs are
subject to a complex set of incentives (e.g., Hellman, 2002) and typically hold large bargaining power
over founders. Empirically, the related evidence is mixed (cf., Masulis and Nahata, 2011; Lindsey,
2008), and scarce, because innovation returns are particularly hard to measure for the private firms that
are the usual targets of VCs. Although alternative interpretations cannot be fully ruled out, additional
results provide supporting evidence that potential returns trickle down to founders. In particular, this
interpretation is supported by results showing that portfolio exchanges also increase in situations where
VCs’ conflicts of interest are potentially low, and where VCs have little bargaining power over the
portfolio companies involved in the exchange with the joiner, such as in mature firms that entered the
portfolio more than five years prior and that VCs are likely to have already exited (see Online Appendix
3).
4.4 Mechanisms
Tables 6 and 7 summarize the results from a battery of tests, which, taken at face value, support
three novel mechanisms of exchanges of innovation resources inside venture capital portfolios.
The first mechanism is spawning, by which I mean instances where entrepreneurs move on to start
new companies also financed by the same VCs. In support of this mechanism, Panel A in Table 6 starts
by showing that inventors that emigrate from the joiner towards the VC’s portfolio appear to go mostly
toward future rather than to incumbent portfolio companies (compare columns 4 and 7 in Panel A). This
21
temporal emigration pattern of inventors is as expected with spawning; the idea is that entrepreneurs
leave to start new companies that have not been founded at the time of the joining event. I classify firms
into incumbent and future portfolio companies according to whether they already exist in the portfolio
during the new joiner’s joining event. Inventor exchanges are then categorized (and denoted by a
corresponding suffix) into future or incumbent according to the type of portfolio company they move
to or from.
Because many founders do not file patents, but most should hold executive positions in the firm, I
provide additional and arguably more direct evidence of the spawning mechanism by investigating the
emigration patterns of the company executives reported in SDC.18 Panel B in Table 6 shows two striking
results in support of the spawning mechanism. First, executive exchanges, as measured by the
movement of executives across companies, are also prevalent inside VC portfolios: there is a 153%
increase in executive exchanges (to and from joiners and the VC’s portfolio) after joiners enter the
portfolio (0.020 relative to the unconditional mean of 0.013; see column 3 in Panel B). Second, company
executives have the same temporal pattern of emigration as inventors: most executives tend to leave for
future, rather than incumbent, portfolio companies (compare columns 4 and 7 in Panel B).19
The second mechanism of exchanges of innovation resources inside venture capital portfolios that
finds support in the data is carve-outs, i.e., instances where entrepreneurs divest some of their innovation
units inside the portfolio. The increase in Relative mergers and acquisitions shown in Table 3 (column
11, Panel B) provides the first piece of evidence consistent with this mechanism. Because carve-outs
are likely to involve the simultaneous exit of several workers (say, all inventors in one same business
unit), rather than the isolated transfer of individual workers, I move on to explore the prevalence of, and
18 I retrieve data on executives from SDC and define an executive as moving from company A to company B at
time t, if at time t the individual appears as an executive of company B, and at any year prior to t the individual
appeared as an executive of company A. The variable Executive emigrates (Executive immigrants) measures the
number of executives that moved from the joiner (any portfolio company) to any of the portfolio companies (the
new joiner) over time. Executive exchanges measure all executive movements. 19 The analysis of executives has some limitations. First, I only capture executive changes if companies raise
rounds of financing. If they do not, they will not appear in the SDC sample. Thus, as with measures of inventor
exchanges, my executive metrics likely underestimate executive exchanges. Second, I cannot construct measures
of relative executive exchanges (which control for technological clustering in portfolios) because there is no
information on the executives of nonportfolio companies.
22
circumstances surrounding, grouped departures. The results in Table 7 provide further support for carve-
outs: Panel A shows that the bulk of inventor emigrations constitute grouped departures (compare
columns 3 and 5); that is, instances in which more than one inventor (i.e., a team of inventors) emigrates.
More importantly, Panel B shows that 7% (3%) of such grouped departures occur within four years of
a within-portfolio merger or acquisition of the joiner by an incumbent (future) portfolio company. Group
departures are hand classified into seven categories of corporate events by comparing the timing of the
emigration and the timing of such corporate events.20 Finally, I make further use of the executives’ data
to provide additional evidence in support of carve-outs. The results in Panel C of Table 6 show that
almost 10% of inventor exchanges coincide with the executives of the joiner also leaving the firm
(column 6).21 This coincidence is consistent with carve-outs, as I expect both scientists and managers
to exit firms in tandem when companies divest some of their innovation units.
The final mechanism of exchange of innovation resources inside venture capital portfolios
supported by the data is recycling, whereby the assets of restructuring portfolio companies are absorbed
by other portfolio firms. Panel B in Table 7 shows that 31% of grouped departures from joiners toward
portfolio companies can be traced to a restructuring event of the new joiner, such as a merger or
acquisition (with a company outside of the portfolio of the VC, 22%), an initial public offering (2%),
or a more informal reorganization (such as the hiring of new CEOs, 7%). Results in the panel also show
that 3% of grouped departures absorbed in the portfolio follow a liquidation event of the new joiner.
Overall, the preponderance of evidence in Tables 6 and 7 supports the mechanisms of spawning,
carve-outs, and recycling, although I cannot directly test any of these channels due to data limitations
(e.g., there are no public records on divesting discussions by VCs and entrepreneurs). These mechanisms
are also broadly consistent with prior work and with anecdotal evidence, but this analysis is, to my
20 The seven categories of corporate events are: merger or acquisition by a company outside the VC portfolio,
merger or acquisition by an incumbent portfolio company, merger or acquisition by a future portfolio company,
IPO, liquidation, restructuring, and founding. The class reorganization includes several types of corporate events
including: the hiring of new CEOs, securing investment from government, and opening a foreign subsidiary. An
inventor emigration is classified as a respective corporate event if it occurs within four years of the event. An
emigration event is unclassifiable if it does not occur within four years of the six corporate events considered. 21 For this complementary test, I combine information on inventors and executives and construct measures of
conditional inventor exchanges (i.e., conditional on simultaneous executive movements).
23
knowledge, their first collective quantitative evidence. For example, spawning echoes the serial
entrepreneur phenomenon studied elsewhere (e.g., Gompers et al., 2010). The novelty here lies in
showing that the recurring pattern is also found in the relations between entrepreneurs and the same
investors rather than just in the unilateral founding decisions of entrepreneurs. While other work
examines the repeated relationships between VCs and serial founders (e.g., Bengtsson, 2013), evidence
on how these repeated relations also extend to inventors is new. Similarly, carve-outs are consistent
with the notion that VCs help companies recruit key personnel (e.g., Hellman and Puri, 2002; Kaplan
and Stromberg, 2001), and support the bridge-building role of VCs in mergers and acquisitions as shown
by Gompers and Xuan (2012). The focus on bulk hiring of inventors is new here, and is reminiscent of
the “acqui-hiring” phenomenon described by the popular press. Finally, recycling is consistent with the
well-documented fact that VCs play a major role in exits, such as structuring mergers, acquisitions,
initial public offerings, and liquidations (e.g., Sahlman, 1990). The evidence on the reuse of human
capital assets inside the portfolio is novel and points to one specific role for VCs: strategic management
of residual assets in restructurings.
5. Exploiting the adoption of PIR across states as an exogenous determinant of VCs’ portfolios
The question remains: if companies were randomly assigned across VCs’ portfolios, would sharing
a common VC facilitate exchanges of innovation resources? Because a large scale randomized trial is
unfeasible in this setting, in this section, I exploit regulatory changes to the investment policy of state
pension funds (namely, the adoption of PIR) as plausible exogenous variation in the composition of
VCs’ portfolios. I begin by providing a background of the PIR. Then, I show that PIR adoption in a
VC’s home state predicts the timing of when joiners enter VCs’ portfolios for the first time as well as
the prevalence of portfolio exchanges between joiners and other portfolio companies. I conclude by
presenting several robustness checks and discussing the interpretation of the results.
5.1 Institutional setting: adoption of prudent investor rules across states
State pension funds are among the most important limited partners in the venture capital industry.
In 2011, they accounted for 28% of new funds committed to venture capital, almost twice the 13%
24
accounted for by the industry’s second most important capital provider, fund of fund managers.22 Prior
work shows that these funds have a substantial home state bias in their private equity investments (see
Hochberg and Rauh, 2013). The likely cause of this home state bias is local development policies, in
the form of economically targeted investments, to which state pension funds are typically subject. For
example, state pension funds are often required to invest in companies and industries located within the
state’s borders as a way of promoting local employment (Brown, Pollet, and Weisbener, 2012).23
State pension funds are governed at the local level. Their significant participation as limited partners
of venture capital firms can be traced back to state-level regulatory changes on such funds’ admissible
investments. In particular, this participation can be linked to the adoption of prudent investor rules (PIR)
that allowed state pension funds, which were not covered in the Employee Retirement Income Security
Act (ERISA) clarification of 1979, to invest in venture capital. State-level PIR adoption was prompted
by the 1994 Uniform Prudent Investor Act (UPIA) of the Uniform Law Commission. Similarly to
ERISA, UPIA requires trustees to become devotees of “modern portfolio theory” and to invest as a
prudent investor would invest by “considering the purposes, terms, distribution requirements, and other
circumstances of the trust” while using “reasonable care, skill, and caution.”
State-level adoption of PIR through UPIA began during the early 1990s, when state pension funds
accounted for only 4% of funds committed to VC.24 Today, the majority of US states (46) have adopted
PIR by enacting UPIA. Adoption was staggered across states, as shown in Fig. 3. Information about
UPIA adoption is from Uniform Laws Annotated, published by Thompson-Reuters and summarized in
Online Appendix 5. I consider two different proxies for the date of PIR adoption in each state: PIR
enactment (i.e., the date the UPIA legislation was signed into law in the state) and PIR implementation
(i.e., the effective date of the UPIA enactment). Online Appendix 5 shows PIR implementation lags
behind PIR enactment in 15 out 46 states. The average lag is 0.37 years among the 46 states and 1.13
22 Source: author’s calculations using Preqin data for 2011. 23 Brown, Pollet, and Weisbener (2012) report, “Overall, we find that the state-managed equity portfolios hold a
broadly diversified portfolio of stocks. Relative to the value weighted index of all US equities, these state managed
plans overweight large (i.e., S&P 500) stocks. […] However, we also find strong evidence that these plans
overweight the stocks of companies that are headquartered in the state.” 24 Source: author’s calculations using Preqin data for 2011.
25
years among the states where there is a lag. Among the early adopters were Illinois, New York, and
California, which adopted PIR via UPIA enactment in 1992, 1994, and 1995, respectively. The latest
adopter of PIR via UPIA enactment was Montana in 2013.
Similar to how ERISA increased funding for the US venture capital industry, state-level PIR
adoption increased local venture capital firms’ funding sources. Following states’ PIR adoption via
UPIA enactment (implementation), capital commitments to local venture capital firms by state pension
funds increased by an average of 175 (187) million USD relative to pension funds located in other states.
The increase was economically significant: it corresponded to a 54% (29%) relative increase over
average nominal (real) capital commitments prior to PIR adoption via UPIA enactment. This increase
was not exclusively driven by California and reflected a general shift in the investment policy of state
pension funds toward local private equity (including buyout funds). To produce these estimates, I
compare changes in capital commitments to local venture capital firms by local state pension funds
(which were affected by the PIR adoption) relative to those from pension funds located elsewhere
(which were not affected by the PIR adoption) after a state’s PIR adoption, using a triple difference–in–
differences methodology. In Online Appendix 6, I explain in detail the data sources and methodology
used to produce these estimates and summarize the results.
5.2 PIR adoption and the timing when joiners enter VC portfolios
The evidence in Section 5.1 suggests that a state’s adoption of PIR provides a source of plausible
exogenous variation in the portfolio composition of local VCs. PIR adoption allows local VCs to make
some investments they otherwise would not have made by increasing their access to funding capital
from local state pension funds.
Consistent with states’ PIR adoption affecting the composition of local VCs’ portfolios, I show in
Fig. 4 that joiners are twice as likely to enter VCs’ portfolios for the first time after PIR adoption in
VCs’ home states. Fig. 4 plots average Post for joiners and VC pairs in the sample against PIR event
time dummies that indicate for each pair the year relative to PIR adoption in the VC’s home state. In
the figure, Panel A (B) dates PIR adoption according to UPIA enactment (implementation); see Section
26
4.1 and Appendix 5. Both panels in the Figure show that roughly ten years prior to PIR adoption, a
slight negative trend exists in average Post. After PIR adoption, however, the trend changes, with
average Post beginning to increase one year after PIR adoption, and continuing to do so for roughly 12
years. The figure also shows a slight upward trend in average Post throughout the entire PIR adoption
event time (see discussion in Section 5.4). To construct this plot, I organize the (joiner VC) pair panel
data in “PIR event time,” where the event is the PIR adoption date in the VC’s home state. I then regress
Post against PIR event time dummies and cluster standard errors at the level of the VC’s home state. I
include the VC’s home state fixed effects in the regression. The solid (dotted) line plots the estimated
coefficients (95th confidence interval) of the PIR event time dummies. The coefficients are normalized
relative to the year before PIR adoption, which is set to zero. Joiner and VC pairs where the VC is
located in a state with no PIR adoption in the sample period are excluded from the plots by design.
I summarize the results of the PIR event study plots in Fig. 4 by estimating the following equation:
𝑃𝑜𝑠𝑡𝑖𝑗𝑡 = 𝛼𝑖𝑗 + 𝛽𝑃𝐼𝑅𝑗𝑡 + 휀𝑖𝑗𝑡 , (3)
where 𝑃𝐼𝑅𝑗𝑡, is an indicator variable that equals one after the state of the VC investor j adopts PIR
on a two-period panel data set (i.e., a pre- and post-PIR adoption period) of the joiner and VC pairs in
the sample for which the home state of the VC adopts PIR within the sample period.25 I collapse the
time series information for each joiner and VC pair to control for potential trends in PIR event time and
serial correlation of standard errors, which are common in difference-in-differences exercises exploiting
the staggered adoption of regulations across states (see Bertrand, Duflo, and Mullainathan, 2004).
Because the timing of PIR adoption across states is not the same, but is instead staggered, to collapse
the time series information, I use residuals from regressing 𝑃𝑜𝑠𝑡𝑖𝑗𝑡, on VC home state fixed effects,
year dummies, and differential trends across the home states of the VCs. For every joiner and VC pair
in the data, the residuals from the years before and after PIR adoption in the VC’s home state are
aggregated, respectively, into the pre- and post-adoption period observations. Eq. (3) is estimated by
25 The states that adopted PIR out of the sample period are Montana, Vermont, Delaware, Kentucky, Louisiana,
and Maryland.
27
using ordinary least squares (OLS), the aggregate residuals as dependent variable, and clustering
standard errors at the level of the home state of the VC.
Column 1 in Panel A of Table 8 shows that the increasing pattern in 𝑃𝑜𝑠𝑡𝑖𝑗𝑡 after PIR adoption that
is shown in Fig. 4 is statistically and economically significant. Joiners are 101% more likely (relative
to the sample mean) to enter the VC’s portfolio after PIR adoption in the VC’s home state (0.618 relative
to 0.61; see Table 2).
5.3 PIR adoption in VCs’ home state and portfolio exchanges
Columns 2-12 in Panel A of Table 8 show that relative portfolio exchanges (between joiners and
other companies in VCs’ portfolios) also increase (namely, they triple) after PIR adoption in the VCs’
home state. The panel presents reduced form estimates of portfolio exchanges on PIR adoption from
estimating Eq. (3) using the aggregate residuals of the different measures of portfolio exchanges, rather
than the aggregate residuals of Post, as dependent variables.26 Across the columns, the point estimates
are positive and statistically significant, except for Relative inventor immigrants in Column 9. Relative
to the sample means, these estimates imply an average increase of 204% in relative portfolio exchanges.
The largest estimated increase is of 425% for Relative citations received (coefficient estimate of 0.170,
see column 2 in Panel A of Table 8) over the sample mean (0.04, see Table 2). The smallest significant
increase is 100% for Relative inventor emigrates (0.010 relative to 0.01, see Table 2).
Under the assumption that PIR adoption in the VCs’ home states does not disproportionately
increase relative portfolio exchanges through channels other than the portfolio joining event, the results
in Panel A of Table 8 suggest that sharing a common VC causally facilitates portfolio exchanges (see
Section 5.4 for a relaxation of this identification assumption). Panel B in Table 8 presents the estimates
of these causal effects using an instrumental variable (IV) estimation: there is an average increase in
relative portfolio exchanges of 329% (relative to the sample means) after joiners enter a VC’s portfolio
26 I aggregate residuals of the different measures of portfolio exchange for every joiner and VC pair using the
same methodology to aggregate residuals for 𝑃𝑜𝑠𝑡𝑖𝑗𝑡 in Section 4.2. Namely, I regress each measure of portfolio
exchange on the VC’s home state fixed effects, year dummies, and differential trends across the home states of
the VCs. Then, for every joiner and VC pair in the data, the residuals from the years before and after PIR adoption
in the VC’s home state are aggregated, respectively, into the pre- and post-adoption period observations.
28
for the first time, which holds for both portfolio exchanges financed by joiners (columns 3, 6, 9) as well
as those financed by portfolio companies (columns 2, 5, 8). Across all columns in Panel B, the IV
estimates are statistically significant for all relative portfolio exchanges except for Relative inventor
immigrants. To produce these IV estimates, I run a two-stage least squares regression of Eq. (2) on the
two-period panel of aggregate residuals, where I instrument the variable Post with the PIR dummy. The
F-test of PIR reported in Panel A (62.76) reveals the instrument’s relevance (see Stock and Yogo, 2005).
In Panel C of Table 8, I present the OLS estimates of Eq. (2) on the two-period panel of aggregate
residuals as a benchmark against which to compare the IV results of Panel B. Relative to the sample
means, the OLS benchmark estimates imply an average increase of 315% in relative portfolio
exchanges.
A comparison between the OLS estimates in Panel C of Table 8 and the OLS estimates in Section
4 (reported in Table 3) reveals that the former are substantially larger. The difference in magnitudes
between these two groups of OLS estimates is due to differences in sample and methodology between
Sections 4 and 5. By construction, the estimates in Table 8 exclude all joiner and VC pairs for which
the state of the VC does not adopt PIR (via UPIA enactment) within the estimation period. In addition,
the methodology in Table 3 allows for both a mean shift and a trend break after the portfolio joining
event. Instead, the methodology behind the estimates in Table 8 does not allow for a trend break,
precisely because the purpose of the methodology is to collapse the time series variation into only two
time periods per joiner and VC pair. To produce estimates of the results in Section 4 that are more
comparable to the ones presented in Table 8, I calculate the implied average increase in portfolio
exchanges from results in Section 4 based on a two-period panel data set that also collapses time series
variation for every joiner and VC pair into two periods: before and after the joining event (see Online
Appendix 3). The implied average increase in relative portfolio exchanges (relative to the mean) based
on these estimates is 123%, much higher than the 60% implied average increase of Table 3, and closer
to the 315% implied average increase of Table 8. One last discrepancy between the methodologies
regards the periods over which the time series variation per joiner and VC pairs is collapsed. In Table
8, this variation is collapsed into pre- and post-PIR adoption periods, whereas in Online Appendix 3, it
29
is collapsed into pre- and post-portfolio-joining-event periods. Since PIR adoption events began only in
the 1990s, this discrepancy can explain the higher magnitude of the results in Table 8. Portfolio joining
events started as early as 1977, and, on average, the incidence of portfolio exchanges (and cross-
company exchanges more generally) is higher after the 1990s than during the 1970s and 1980s.
In addition, a comparison of estimates across Panels B and C in Table 8 reveals that IV estimates
are, on average, 4% larger than OLS estimates. This positive difference is not statistically significant,
but it makes the interpretation of the results difficult (i.e., one cannot determine how much of the OLS
estimate is selection and how much it is treatment), as is common in the venture capital literature (e.g.,
Kortum and Lerner, 2000; Mollica and Zingales, 2007; Hirukawa and Ueda, 2008; Nanda and Rhodes-
Kropf, 2013; Popov and Rosenboom, 2011; see Online Appendix 7).
One potential explanation for the positive difference is that marginal joiners face high costs in
attracting VCs absent PIR adoption; for example, Nanda and Rhodes-Kropf (2013) show that VCs
usually experiment whenever they have an influx of funds by investing in riskier companies. In that
case, IV would yield a “local average treatment effect” above the average marginal effect in the
population and potentially above OLS (Angrist and Imbens, 1994). Other potential explanations
include measurement error from differences between the enactment and actual implementation dates
of PIR (cf., Card, 2001). Against this explanation, the IV estimate also exceeds the OLS estimate when
I proxy PIR adoption using UPIA implementation dates (see Online Appendix 8).
5.4. Robustness checks
In this section, I discuss several potential methodological concerns with using PIR adoption as an
exogenous determinant of VCs’ portfolios and present suggestive evidence against them.
The main methodological concern with the results in Table 8 is that PIR adoption is endogenous to
the innovation opportunities of joiners and portfolio companies. That is, PIR adoption anticipates (or
leads to venture capital unrelated) changes in innovation opportunities for joiners and portfolio
companies. Three additional sets of findings provide evidence against the relative importance of this
concern. First, Online Appendix 8 shows no significant evidence of correlated trends between relative
30
portfolio exchanges and PIR adoption, although weak increases are visible three years before adoption
for some of the measures. Second, Panel A of Table 9 shows similar results when I restrict the sample
to pairs where the joiner and the portfolio companies are not in the home state of the VC.27 This second
set of results point to causal effects as long as PIR adoption in the VC’s home state affects relative
exchanges between the out-of-state joiners and out-of-state portfolio companies only through the joining
event. In support of this assumption, Online Appendix 8 shows no significant evidence of correlated
trends in PIR adoption and measures of portfolio exchange in this subsample. The last set of findings
mitigating PIR endogeneity concerns show similar results when I further restrict the sample to pairs of
joiners and portfolio companies that are also not in coincidental states where PIR is adopted at the same
time as in the home state of the VC (see Panel B of Table 9).
A second methodological concern is that the results in Table 8 are driven by aggregate trends that
are visible in the plots of average Post against PIR event time (Fig. 4; see discussion in Section 5.1). To
address this concern, I restrict the sample to the joiner and VC pairs for which the entire 11-year
portfolio joining event time period is within a 20-year interval around the PIR adoption of the VC’s
home state. Other joiner and VC pairs may have financing events that are too far from the PIR adoption
event and are likely to add noise rather than help reduce any potential bias in the OLS estimates. The
results are summarized in Online Appendix 8. They show that aggregate trends do not drive the results
in Table 8: the reduced form, IV, and OLS estimates are similar for the restricted sample.
A third concern regards the potential driving effect of California or Massachusetts. Against it,
Online Appendix 8 shows that results are similar when I drop California and/or Massachusetts from the
analysis sample, except for all proxies of inventor exchanges. For these measures, none of the estimates
(reduced form, OLS, or IV) are statistically significant. One potential explanation is the enforcement of
noncompete agreements imposed in other states during some intervals of the sample period (see Marx,
27 Only results for inventor-based exchanges do not hold. One interpretation of these results is that the anticipation
and facilitation by VCs of “human capital related” portfolio exchanges requires closer monitoring, such as the
level of monitoring imposed by VCs that are geographically close to their companies (see Lerner, 1995; Bernstein,
Giroud, and Townsend, 2016).
31
2013). These noncompete agreements restrict the movement of workers across companies and can
impede the role of VCs in anticipating and facilitating human capital reallocation inside their portfolios.
A final concern regards potential measurement error in the actual adoption dates of PIR across
states, as the average lag between PIR adoption via UPIA enactment and via UPIA implementation is
0.37 years (see Section 5.1). Against this concern, Online Appendix 8 shows that the results are similar
if I date PIR adoption using UPIA implementation rather than UPIA enactment (see Online Appendix
5).
6. Conclusion
I show that several proxies of exchanges of innovation resources are prevalent between companies
sharing common venture capital investors. Results hold for companies that secure venture capital for
plausibly exogenous reasons and in situations where VCs’ bargaining power and potential conflicts of
interest are low. The data support three novel mechanisms of resource exchange inside portfolios:
carve-outs, spawning, and recycling, whereby entrepreneurs divest innovation units, start new
ventures, and reuse residual assets in other portfolio companies, respectively. As the successful
commercialization of inventions typically requires these cross-company exchanges, the results suggest
that returns to innovation are higher inside VC portfolios—although not necessarily appropriated by
the original inventors (VCs face a complex set of incentives and hold strong bargaining power over
founders). The results help explain why venture capital disproportionately contributes to innovation
relative to other financing sources such as corporate investment (Kortum and Lerner, 2000).
32
References
Akcigit, U., Celik, M., Greenwood, J., 2016. Buy, keep or sell: economic growth and the market for
ideas. Econometrica 84, 943-984.
Agrawal, A., Cockburn, I., McHale, J., 2006. Gone but not forgotten: knowledge flows, labor
mobility, and enduring social relationships. Journal of Economic Geography 6, 571-591.
Allen, J., Philips, G., 2000. Corporate equity ownership, strategic alliances, and product market
relationships. Journal of Finance 55, 2791-2815.
Almeida, P., Kogut, B., 1999. Localization of knowledge and the mobility of engineers in regional
networks. Management Science 45, 905-917.
Angrist, J., Imbens, G., 1994. Identification and estimation of local average treatment effects.
Econometrica 62, 467-475.
Arrow, K. J., 1975. Vertical integration and communication. Bell Journal of Economics, RAND
Corporation 6, 173-183.
Azar, J., Schmalz, M., Tecu, I., 2018. Anti-competitive effects of common ownership. Journal of
Finance 73, 1513-1565.
Azoulay, P., Graff Zivin, J., Sampat, B., 2012. The diffusion of scientific knowledge across time and
space: evidence from professional transitions for the superstars of medicine. In: Lerner, J., Stern, S.
(Eds.), The Rate and Direction of Inventive Activity: A New Agenda. University of Chicago Press,
Chicago, pp. 107-155.
Bengtsson, O., 2013. Relational venture capital financing of serial founders. Journal of Financial
Intermediation 22, 308-334.
Bernstein, S., Lerner, J., Sørensen, M., Stromberg, P., 2016. Private equity and industry performance.
Management Science 63, 1198-1213.
Bernstein, S., Giroud, X., Townsend, R., 2016. Journal of Finance 71, 1591-1622.
Bernstein, J., Nadiri, M. I., 1989. Research and development and intra-industry spillovers: an
empirical application of dynamic duality. Review of Economic Studies 56, 249-269.
Bertrand, M., Duflo, E., Mullainathan, S., 2004. How much should we trust difference-in-differences
estimates? The Quarterly Journal of Economics 119, 249-275.
Branstetter, L. G., Sakakibara, M., 2002. When do research consortia work well and why? Evidence
from Japanese panel data. American Economic Review 92, 143-159.
Branstetter, L., 2006. Is foreign direct investment a channel of knowledge spillovers: evidence from
Japan’s FDI in the United States. Journal of International Economics 68, 325-344.
Brown, J. R., Pollet, J. M., Weisbenner, S. J., 2012. The investment behavior of state pension plans.
Unpublished working paper. NBER.
Card, D., 2001. Estimating the return to schooling: progress on some persistent econometric
problems. Econometrica 69, 1127–1160.
Chan, S., Kensinger, J., Keown, A., Martin, J., 1997. Do strategic alliances create value? Journal of
Financial Economics 46, 199-221.
Denis, D. J., and Denis, D. K. 1995. Performance changes following top management dismissals.
The Journal of Finance 50, 1029-1057.
33
Fallick, B., Fleischman, C., and Rebitzer, J., 2006. Job-hopping in Silicon Valley: some evidence
concerning the microfoundations of a high-technology cluster. Review of Economics and Statistics 88,
472-481.
Fulghieri, P., Sevilir, M., 2009. Size and focus of a venture capitalist's portfolio. Review of
Financial Studies 22, 4643-4680.
Galasso, A., Schankerman, M., Serrano, C. J., 2013. Trading and enforcing patent rights. RAND
Journal of Economics 44, 275-312.
Gomes-Casseres, B., Hagedoom, J., Jaffe, A. B., 2006. Do alliances promote knowledge flows?
Journal of Financial Economics 80, 5-33.
Gompers, P., Xuan, Y., 2012. The role of venture capitalists in the acquisition of private companies,
In: Barth, J. R., Lin, C., Wihlborg, C. (Eds.). Research Handbook on International Banking and
Governance. Edward Elgar, Cheltenham, pp. 535-565.
Gompers, P., Kovner, A., Lerner, J., Scharfstein, D., 2010. Performance persistence in
entrepreneurship. Journal of Financial Economics 96, 18-32.
González-Uribe, J., 2013. Venture capital and innovation. Unpublished working paper. Columbia
University.
Hall, B., Jaffe, A., Trajtenberg, M., 2001. The NBER patent citations data file: lessons, insights and
methodological tools. Unpublished working paper. NBER.
Hall, B., Jaffe, A., Trajtenberg, M., 2005. Market value and patent citations. Rand Journal of
Economics 36, 16-38.
Hellmann, T., 2002. A theory of strategic venture investing. Journal of Financial Economics 64, 285-
314.
Hellmann, T., Puri, M., 2002. Venture capital and the professionalization of start-up firms: empirical
evidence. Journal of Finance 57, 169-197.
Hirukawa, M., Ueda, M., 2008. Venture capital and industrial innovation. Unpublished working
paper. University of Wisconsin, Madison and Setsunan University.
Hochberg, Y., Lindsey, L., Westerfield, M., 2015. Resource accumulation through economic ties:
evidence from venture capital. Journal of Financial Economics 118, 245-267.
Hochberg, Y., Rauh, J., 2013. Local overweighting and underperformance: evidence from limited
partner private equity investments. Review of Financial Studies 26, 403-451.
Hochberg, Y., Mazzeo, M., McDevitt, R., 2015. Specialization and competition in the venture capital
industry. Review of Industrial Organization 46, 323-347.
Hochberg, Y., Serrano, C. J., Ziedonis, R. H., 2018. Patent collateral, investor commitment, and the
market for venture lending. Journal of Financial Economics 130, 74-94.
Jaffe, A. B., Trajtenberg, M., Henderson, M., 1993. Geographic localization of knowledge
spillovers as evidenced by patent citations. Quarterly Journal of Economics 108, 577-598.
Kaiser, U., Kongsted, H. C., Rønde, T., 2015. Does the mobility of R&D labor increase
innovation? Journal of Economic Behavior and Organization, Elsevier 110, 91-105.
Kaplan, S. N., Lerner, J., 2017. Venture capital data: opportunities and challenges. In: Haltiwanger,
J., Hurst, E., Miranda, J., Schoar, A. (Eds.), Measuring Entrepreneurial Businesses: Current
Knowledge and Challenges. University of Chicago Press, Chicago, pp. 413-431.
34
Kaplan, S. N., Stromberg, P., 2001.Venture capitals as principals: contracting, screening, and
monitoring. American Economic Review 91, 426-430.
Kaplan, S. N., Stromberg, P., 2002. Financial contracting theory meets the real world: an empirical
analysis of venture capital contracts. Review of Economic Studies, 1-35.
Kaplan, S. N., 2006, Mergers and acquisitions: a financial economics perspective. Unpublished
working paper. University of Chicago.
Kaplan, S.N., Stromberg, P., 2009. Leveraged buyouts and private equity. Journal of Economic
Perspectives 23, 121-146.
Kogan, L., Papanikolaou, D., Seru, A., Stoffman, N., 2017. Technological innovation, resource
reallocation and growth. Quarterly Journal of Economics 132, 665-712.
Kortum, S., Lerner, J., 2000. Assessing the contribution of venture capital to innovation. RAND
Journal of Economics 31, 674-692.
Lai, R., D’Amour A., Fleming, L., 2009. The careers and coauthorship networks of U.S. patent-
holders, since 1975. Unpublished working paper. Harvard Business School.
Lerner, J., 1995. Venture capital and the oversight of private firms. Journal of Finance 50, 301-318.
Lerner, J., 2009. Boulevard of broken dreams: why public efforts to boost entrepreneurship and
venture capital have failed—and what to do about it. Princeton University Press, Princeton.
Lindsey, L., 2008. Blurring firm boundaries: The role of venture capital in strategic alliances. Journal
of Finance 63, 1137-1168.
Maats, F., Metrick, A.,Yasuda, A., Hinkes, B., Vershovski, S., 2011. On the consistency and
reliability of venture capital databases. Unpublished working paper. University of California at Davis.
Mann, W., 2018. Creditor rights and innovation: evidence from patent collateral. Journal of Financial
Economics 130, 25-47.
Masulis, R. W., Nahata, R., 2011. Venture capital conflicts of interest: evidence from acquisitions
of venture backed firms. Journal of Financial and Quantitative Analysis 46, 395-430.
Marx, M., 2013. The firm strikes back: Non-compete agreements and the mobility of technical
professionals. American Sociological Review 6, 695-712.
Mollica, M. A., Zingales, L., 2007. The impact of venture capital on innovation and the creation of
new business. Unpublished working paper. University of Chicago.
Mowery, D., Oxley, J., Silverman, B., 1996. Strategic alliances and interfirm knowledge transfer.
Strategic Management Journal 17, 77-91.
Nanda, R., Rhodes-Kropf, M., 2013. Investment cycles and startup innovation. Journal of Financial
Economics 110, 403–418.
Popov, A., Rosenboom, P., 2009. Does private equity investment spur innovation? Evidence from
Europe. Unpublished working paper. European Central Bank.
Roach, M., Cohen, W., 2013. Lens or prism? Patent citations as a measure of knowledge flows
from public research. Management Science 59, 504-525.
Rosenberg, N., 1972. Factors affecting the diffusion of technology. Explorations in Economic History
10, 3-33.
Sahlman, W., 1990. The structure and governance of venture capital organizations. Journal of
Financial Economics 27, 473-521.
35
Serrano, C., 2010. The dynamics of the transfer and renewal of patents. RAND Journal of Economics
41, 686-708.
Seru, A., 2014. Firm boundaries matter: Evidence from conglomerates and R&D activity. Journal of
Financial Economics 111, 381-405.
Sørensen, M., 2007. How smart is smart money? A two-sided matching model of venture capital.
Journal of Finance 62, 2725-2762.
Spence, M., 1984. Cost reduction, competition, and industry performance. Econometrica 52, 101-
122.
Stock J., Yogo, M. 2005. Testing for weak instruments in linear IV regression. In: Andrews D. W.
K., Stock, J. H. (Eds.) Identification and Inference for Econometric Models. Cambridge University
Press, New York, pp. 80-108.
Stuart, T., 2000. Interorganizational alliances and the performance of firms: a study of growth and
innovation rates in a high-technology industry. Strategic Management Journal 21, 791-811.
Teece, D., 1980. Economies of scope and the scope of the enterprise. Journal of Economic Behavior
and Organization 1, 223-247.
Teece, D., 1994. Profiting from technological innovation: Implications for integration, collaboration,
licensing and public policy. Transnational Corporations and Innovatory Activities 15, 73.
Teece, D., 2010. Technological innovation and the theory of the firm. In: Hall, B. H., Rosenberg, N.
(Eds.) Handbook of the Economics of Innovation, Vol. 1. Elsevier, Amsterdam, pp. 679-730.
Thompson, P., Fox-Kean, M., 2005. Patent citations and the geography of knowledge spillovers: A
reassessment. American Economic Review 95, 450-460.
36
Fig. 1 Patterns in portfolio exchanges around portfolio joining events
Panel A—Citations received Panel B—Citations made Panel C—Overall citations
Panel D—Patents sold Panel E—Patents bought Panel F—Patent sales
Panel G—Inventor emigrates Panel H—Inventor immigrants Panel I—Inventor exchanges
Panel J—Alliances Panel K—Mergers and acquisitions
The figure plots the 𝛽𝜏 coefficients on the dummies for years relative to the venture capital portfolio joining event from estimating
regression (1) for the different proxies of portfolio exchanges. The event window is five years before and six years after the joiner enters
the VC’s portfolio for the first time. The dashed lines show the 95% confidence interval on each coefficient; standard errors are clustered
at the joiner and VC pair level. Year -1 is the year before the joiner enters the VC’s portfolio for the first time. The coefficients are
normalized relative to year -1, which is set to zero. All the regressions include joiner and VC pair fixed effects and calendar year fixed
effects. An observation is a joiner and VC pair cross time (year).
37
Fig. 2 Patterns in relative portfolio exchanges around portfolio joining events
Panel A—Relative citations received Panel B—Relative citations made Panel C—Relative overall citations
Panel D—Relative patents sold Panel E—Relative patents bought Panel F—Relative patent sales
Panel G—Relative inventor emigrates Panel H—Relative inventor immigrants Panel I—Relative inventor exchanges
Panel J— Relative alliances Panel K—Relative mergers and acquisitions
The figure plots the 𝛽𝜏 coefficients on the dummies for years relative to the venture capital portfolio joining event from estimating
regression (1) for the different proxies of relative portfolio exchanges. The event window is five years before and six years after the
joiner enters the VC’s portfolio for the first time. The dashed lines show the 95% confidence interval on each coefficient; standard errors
are clustered at the joiner and VC pair level. Year -1 is the year before the joiner enters the VC’s portfolio for the first time. The
coefficients are normalized relative to year -1, which is set to zero. All the regressions include joiner and VC pair fixed effects and
calendar year fixed effects. An observation is a joiner and VC pair cross time (year).
38
Fig. 3 PIR adoption across US states
The figure plots the number of US states adopting PIR across time. The red dotted bars indicate years of PIR enactment, as indicated by
the corresponding UPIA legislation dates (see Online Appendix 5). The striped black bars indicate the years of the final implementation
of PIR, as indicated by the dates in which the corresponding UPIA laws became effective (see Online Appendix 5).
39
Fig. 4 Probability of joining the VC’s portfolio around the year of PIR adoption in VC’s home state
Panel A—Probability of joining the VC’s portfolio around PIR enactment in VC’s home state
Panel B—Probability of joining the VC’s portfolio around PIR implementation in VC’s home state
The figure plots the coefficients on the dummies for years relative to the time of PIR adoption in the VC’s home state (PIR event time
dummies), from regressing Post (i.e., a dummy equal to one after the Joiner first enters the VC’s portfolio) against the PIR event time
dummies. The event window is 29 years before and 16 years after the adoption of PIR in the home state of the VC. The solid (dotted)
line plots the estimated coefficients (95th confidence interval) of the PIR event time dummies. The coefficients are normalized relative
to the year before PIR adoption, which is set to zero. All the regressions include VC’s home state fixed effects. An observation is a joiner
and VC pair cross year. Panel A uses information on PIR adoption via enactment (i.e., when the law was approved by the local
government), and Panel B on PIR adoption via implementation (i.e., when the law became effective); see Online Appendix 5 and Fig. 3.
Joiner and VC pairs where the VC is located in a state with no PIR adoption in the sample are excluded from the plots by design.
40
Table 1
Sample composition
Panel A. Distribution over time of venture capital portfolio joining events and patent awards
Venture capital portfolio joining events Patent awards
Total Seed Early stage Expansion Later stage Patent filings Patent grants
1976 10 7
1977 6 5 1 0 0 26 10
1978 12 10 0 2 0 28 19
1979 23 9 7 7 0 44 27
1980 86 55 15 15 1 51 40
1981 219 121 49 42 7 124 33
1982 260 103 66 69 22 166 89
1983 348 156 77 87 28 197 154
1984 336 159 79 81 17 268 184
1985 341 134 62 115 30 340 239
1986 290 132 58 66 34 442 380
1987 275 134 67 57 17 558 427
1988 270 114 77 66 13 718 660
1989 343 115 94 102 32 834 679
1990 221 78 43 83 17 1,026 802
1991 165 39 49 61 16 1,170 1,002
1992 184 54 39 72 19 1,407 1,161
1993 200 78 58 46 18 1,730 1,356
1994 216 77 60 44 35 2,357 1,560
1995 278 97 97 64 20 3,816 2,006
1996 244 71 61 91 21 4,066 2,493
1997 343 75 123 99 46 5,066 3,985
1998 571 140 205 169 57 5,252 4,309
1999 810 122 269 321 98 5,867 4,710
2000 1,171 71 480 473 147 6,918 5,088
2001 856 73 330 362 91 7,640 5,385
2002 721 46 247 304 124 7,558 6,013
2003 807 52 260 299 196 5,889 5,888
2004 743 43 192 278 230 5,118 5,464
2005 559 23 144 203 189 3,708 7,159
2006 395 22 102 146 125 1,836 6,769
2007 335 24 63 124 124 516 6,673
2008 187 12 39 71 65 25 7
Total 11,815 2,390 1,278 288 1,747 74,771 74,771
Panel B. Distribution of patents and joiners across states (Top 10 states)
Patents % Patents Joiners % Joiners
Venture
capital firms
% Venture
capital firms
CA 39,393 56.04 1,845 46.59 415 32.81
MA 6,913 9.83 517 13.06 139 10.99
TX 5,056 7.19 191 4.82 60 4.74
CO 2,589 3.68 110 2.78 32 2.53
WA 2,098 2.98 121 3.06 33 2.61
PA 1,115 1.59 111 2.8 35 2.77
NJ 1,022 1.45 106 2.68 24 1.9
NY 993 1.41 100 2.53 130 10.28
IL 966 1.37 71 1.79 46 3.64
MN 864 1.23 63 1.59 28 2.21
Panel C. Distribution of patents across 2-digit technology classes (Top 10 classes)
Technology class Name Number % Patents
22 Computer Hardware and Software 13,000 17.39
21 Communications 8,142 10.89
46 Semiconductor Devices 7,159 9.57
41
32 Surgery & Medical Instruments 6,589 8.81
24 Information Storage 5,818 7.78
31 Drugs 4,736 6.33
33 Biotechnology 3,753 5.02
41 Electrical Devices 3,302 4.42
19 Miscellaneous Chemical 2,724 3.64
45 Power Systems 2,266 3.03
Panel D. Industry distribution of joiners
Industries Number of joiners % of joiners
Medical health 699 17.65
Semiconductors 659 16.64
Computer software 649 16.39
Communications and media 446 11.26
Biotechnology 432 10.91
Internet specific 363 9.17
Computer hardware 342 8.64
Industrial energy 250 6.31
Other products 67 1.69
Consumer related 53 1.34
Total 3,960 100
The final sample includes information for 3,960 joiners that filed 74,771 patents (approximately 2.5% of all patent awards) and first
entered the portfolio of one of 1,265 VCs during 1976-2008. The data include 11,815 portfolio joining events. All additional financing
rounds by an existing VC investor are not counted as portfolio joining events. In Panel B, the joiner’s home state is retrieved from the
SDC data set. For some joiners, this information is missing. I used web searches based on the joiner’s name to fill in the missing
information. For 649 joiners, the web search returned no information, most likely because they are no longer active. In very few cases,
the joiner’s home state differs from the location of the inventor reported in the patent filings (<5%). In Panel C, I reclassified the three-
digit technology classes of patents into two-digit classes using the method by Hall, Jaffe, and Trajtenberg (2001). In Panel D, I retrieve industry classification from the SDC data set.
42
Table 2
Summary statistics
Observations Mean Std. Dev. Min Max
Citations received 121,553 0.05 1.50 0.00 263.00
Citations made 121,553 0.10 2.35 0.00 249.00
Overall citations 121,553 0.15 2.81 0.00 263.00
Patents sold 121,553 0.01 0.23 0.00 41.00
Patents bought 121,553 0.01 0.24 0.00 22.00
Patent sales 121,553 0.02 0.39 0.00 41.00
Inventor emigrates 121,553 0.01 0.12 0.00 8.00
Inventor immigrants 121,553 0.01 0.17 0.00 12.00
Inventor exchanges 121,553 0.02 0.21 0.00 12.00
Alliances 121,553 0.001 0.034 0.00 1.00
Mergers and acquisitions 121,553 0.002 0.043 0.00 1.00
Matched citations received 121,553 0.01 0.56 0.00 120.00
Matched citations made 121,553 0.03 0.74 0.00 119.00
Matched overall citations 121,553 0.04 0.97 0.00 132.00
Matched patents sold 121,553 0.00 0.02 0.00 4.00
Matched patents bought 121,553 0.00 0.08 0.00 20.00
Matched patent sales 121,553 0.00 0.08 0.00 20.00
Matched inventor emigrates 121,553 0.00 0.11 0.00 30.00
Matched inventor immigrants 121,553 0.00 0.11 0.00 30.00
Matched inventor exchanges 121,553 0.00 0.11 0.00 30.00
Matched alliances 121,553 0.000 0.008 0.00 1.00
Matched mergers and acquisitions 121,553 0.000 0.000 0.00 0.00
Relative citations received 121,553 0.04 1.59 -120.00 263.00
Relative citations made 121,553 0.07 2.39 -104.00 249.00
Relative overall citations 121,553 0.11 2.91 -131.00 262.00
Relative patents sold 121,553 0.01 0.23 -4.00 41.00
Relative patents bought 121,553 0.01 0.25 -20.00 22.00
Relative patent sales 121,553 0.02 0.40 -20.00 41.00
Relative inventor emigrates 121,553 0.01 0.16 -30.00 8.00
Relative inventor immigrants 121,553 0.01 0.20 -30.00 12.00
Relative inventor exchanges 121,553 0.01 0.23 -30.00 12.00
Relative alliances 121,553 0.001 0.035 -1.00 1.00
Relative mergers and acquisitions 121,553 0.002 0.043 0.00 1.00
Post 121,553 0.61 0.49 0.00 1.00
Year 121,553 1997 7.85 1976 2008
Year financing 121,553 1996 7.54 1977 2008
Young company 120,610 0.59 0.49 0.00 1.00
Early stage 121,553 0.43 0.49 0.00 1.00
Junior VC 111,383 0.51 0.50 0.00 1.00
Specialist VC 116,452 0.10 0.29 0.00 1.00
This table presents the summary statistics of the main variables used in the empirical analysis. Observations are at the joiner and VC pair
cross time (year). Section 3 has a detailed explanation of the variables.
43
Table 3
Portfolio exchanges between joiners and portfolio companies
Panel A—Portfolio exchanges
Panel B—Relative portfolio exchanges
This table reports the coefficients and standard errors (in parentheses) from estimating Eq. (2) on the different proxies of portfolio exchanges. An observation is a joiner and VC pair cross time (year). The
explanatory variables of interest are Post, an indicator variable that equals one after the joiner enters the VC portfolio for the first time, and Post×Event-Trend, which equals zero before the joiner enters the
VC portfolio for the first time and indicates the first through sixth years after the joiner enters the portfolio. Standard errors are heteroskedasticity robust and clustered at the joiner and VC pair level.*, **,
and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Dependent variable Citations
received
Citations
made
Overall
citations Patents sold
Patents
bought Patent sales
Inventor
emigrates
Inventor
immigrants
Inventor
exchanges Alliances
Mergers and
acquisitions
Post 0.021*** 0.100*** 0.121*** 0.005*** 0.009*** 0.014*** 0.003** 0.007*** 0.011*** 0.000 0.000
(0.006) (0.022) (0.023) (0.002) (0.002) (0.003) (0.001) (0.002) (0.003) (0.000) (0.000)
Post× Event-Trend 0.033*** -0.026** 0.007 0.002*** 0.001 0.003*** -0.001 -0.006*** -0.005*** 0.001*** 0.001***
(0.005) (0.011) (0.012) (0.001) (0.001) (0.001) (0.000) (0.001) (0.001) (0.000) (0.000)
Obs. 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553
R-2 0.211 0.312 0.293 0.183 0.196 0.222 0.587 0.633 0.605 0.476 0.422
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Dependent variable Relative
Citations
Received
Relative
Citations
Made
Relative
Overall
Citations
Relative
Patents Sold
Relative
Patents
Bought
Relative
Patent Sales
Relative
Inventor
Emigrates
Relative
Inventor
Immigrants
Relative
Inventor
Exchanges
Relative
Alliances
Relative
Mergers and
Acquisitions
Post 0.018*** 0.060*** 0.078*** 0.004*** 0.009*** 0.014*** 0.005*** 0.008*** 0.012*** 0.000 0.000
(0.007) (0.022) (0.023) (0.002) (0.002) (0.003) (0.002) (0.002) (0.003) (0.000) (0.000)
Post×Event-Trend 0.024*** -0.025** -0.001 0.002*** 0.001 0.004*** -0.001* -0.006*** -0.005*** 0.001*** 0.001***
(0.005) (0.011) (0.012) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553
R-2 0.196 0.291 0.267 0.182 0.187 0.217 0.517 0.591 0.571 0.471 0.422
44
Table 4
Heterogeneity I: Relative portfolio exchanges between joiners and portfolio companies
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Dependent variable Relative
citations
received
Relative
citations
made
Relative
overall
citations
Relative
patents
sold
Relative
patents
bought
Relative
patent sales
Relative
Inventor
Emigrates
Relative
inventor
immigrants
Relative
inventor
exchanges
Relative
alliances
Relative
mergers and
acquisitions
Panel A—Age joiner
Young
Post 0.019** 0.116*** 0.135*** 0.002 0.008** 0.010** 0.005** 0.013*** 0.017*** 0.000 -0.001***
(0.008) (0.035) (0.037) (0.002) (0.003) (0.004) (0.002) (0.003) (0.004) (0.000) (0.000)
Post× Event-Trend 0.028*** -0.033* -0.005 0.002*** 0.001 0.003** -0.000 -0.008*** -0.006*** 0.001*** 0.002***
(0.006) (0.018) (0.019) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 71,181 71,181 71,181 71,181 71,181 71,181 71,181 71,181 71,181 71,181 71,181
R-2 0.206 0.293 0.282 0.180 0.164 0.192 0.181 0.164 0.187 0.465 0.349
Old
Post 0.016 -0.025 -0.009 0.009*** 0.012*** 0.021*** 0.005** 0.002 0.006 -0.000 0.001**
(0.012) (0.018) (0.021) (0.003) (0.004) (0.005) (0.003) (0.003) (0.004) (0.000) (0.001)
Post× Event-Trend 0.020** -0.013 0.007 0.004** 0.003 0.007** -0.002** -0.003** -0.003** 0.001*** 0.001**
(0.009) (0.012) (0.016) (0.002) (0.002) (0.003) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 49,429 49,429 49,429 49,429 49,429 49,429 49,429 49,429 49,429 49,429 49,429
R-2 0.185 0.280 0.228 0.182 0.207 0.233 0.135 0.161 0.165 0.481 0.478
p-value F-tests
Post 0.84 0.00 0.00 0.04 0.44 0.12 0.79 0.02 0.04 0.67 0.00
Post× Event-Trend 0.45 0.33 0.62 0.32 0.25 0.23 0.12 0.01 0.08 0.95 0.12
Panel B—Stage joiner
Early
Post 0.003 0.061** 0.063** 0.005* 0.009** 0.014*** 0.006*** 0.017*** 0.021*** -0.000 -0.001**
(0.005) (0.028) (0.029) (0.003) (0.004) (0.005) (0.002) (0.004) (0.004) (0.000) (0.000)
Post× Event-Trend 0.018*** -0.001 0.017* 0.002 0.001 0.003 -0.000 -0.008*** -0.006*** 0.001*** 0.002***
(0.005) (0.009) (0.010) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 51,803 51,803 51,803 51,803 51,803 51,803 51,803 51,803 51,803 51,803 51,803
R-2 0.181 0.286 0.274 0.241 0.209 0.251 0.167 0.158 0.179 0.361 0.329
Late
Post 0.027*** 0.060* 0.088** 0.004* 0.009*** 0.013*** 0.004* 0.002 0.005 0.000 0.001
(0.011) (0.033) (0.035) (0.002) (0.003) (0.004) (0.002) (0.003) (0.003) (0.000) (0.000)
Post× Event-Trend 0.027*** -0.041** -0.014 0.003*** 0.001 0.004** -0.002** -0.005*** -0.004*** 0.001*** 0.001**
(0.008) (0.018) (0.020) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 69,750 69,750 69,750 69,750 69,750 69,750 69,750 69,750 69,750 69,750 69,750
R-2 0.198 0.293 0.266 0.150 0.170 0.187 0.149 0.167 0.177 0.524 0.471
45
This table reports the coefficients and standard errors (in parentheses) from estimating Eq. (2) on the different proxies of relative portfolio exchanges. An observation is a joiner and VC pair cross time
(year). The explanatory variables of interest are Post, an indicator variable that equals one after the joiner enters the VC portfolio for the first time, and Post×Event-Trend, which equals zero before the
p-value F-tests
Post 0.03 0.99 0.59 0.70 0.96 0.88 0.56 0.00 0.00 0.21 0.01
Post× Event-Trend 0.36 0.04 0.16 0.36 0.89 0.66 0.09 0.05 0.22 0.40 0.01
Panel C—Joiner is new to venture capital
New to Venture Capital
Post 0.020** 0.086** 0.106*** 0.004 0.008** 0.012*** 0.005** 0.012*** 0.015*** 0.000 -0.001**
(0.009) (0.034) (0.036) (0.002) (0.003) (0.005) (0.002) (0.003) (0.004) (0.000) (0.000)
Post× Event-Trend 0.025*** -0.022* 0.003 0.003*** 0.002 0.005*** 0.000 -0.006*** -0.005*** 0.001*** 0.001***
(0.006) (0.012) (0.013) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 71,077 71,077 71,077 71,077 71,077 71,077 71,077 71,077 71,077 71,077 71,077
R-2 0.211 0.278 0.270 0.183 0.189 0.223 0.168 0.168 0.184 0.428 0.367
Already Secured Venture Capital
Post 0.018 0.024 0.042 0.005*** 0.011*** 0.016*** 0.005* 0.004 0.008** -0.000 0.001
(0.012) (0.025) (0.027) (0.002) (0.003) (0.004) (0.003) (0.003) (0.004) (0.000) (0.001)
Post× Event-Trend 0.024** -0.030 -0.007 0.001 0.001 0.002 -0.003** -0.006*** -0.005*** 0.001*** 0.001**
(0.010) (0.021) (0.023) (0.001) (0.001) (0.002) (0.001) (0.001) (0.002) (0.000) (0.000)
Obs. 50,476 50,476 50,476 50,476 50,476 50,476 50,476 50,476 50,476 50,476 50,476
R-2 0.181 0.317 0.263 0.182 0.185 0.201 0.146 0.160 0.170 0.509 0.466
p-value F-tests
Post 0.88 0.14 0.15 0.60 0.58 0.50 0.98 0.08 0.15 0.74 0.03
Post× Event-Trend 0.92 0.73 0.72 0.15 0.51 0.21 0.03 0.71 0.96 0.91 0.09
Panel D—VC experience
Junior
Post 0.011 -0.001 0.010 0.006*** 0.010*** 0.016*** 0.003 0.002 0.005 0.000 -0.000
(0.007) (0.007) (0.010) (0.002) (0.003) (0.004) (0.002) (0.003) (0.003) (0.000) (0.000)
Post× Event-Trend 0.013*** -0.000 0.013** 0.003* 0.003* 0.005** 0.000 -0.002* -0.000 0.001*** 0.001*
(0.005) (0.002) (0.005) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 57,344 57,344 57,344 57,344 57,344 57,344 57,344 57,344 57,344 57,344 57,344
R-2 0.224 0.277 0.254 0.234 0.288 0.294 0.143 0.142 0.167 0.476 0.387
Senior
Post 0.028** 0.140*** 0.168*** 0.003 0.010** 0.013** 0.009*** 0.019*** 0.025*** -0.000 -0.000
(0.014) (0.051) (0.053) (0.003) (0.004) (0.006) (0.003) (0.005) (0.005) (0.000) (0.001)
Post× Event-Trend 0.033*** -0.059*** -0.026 0.003** 0.000 0.003 -0.002** -0.011*** -0.011*** 0.001*** 0.002***
(0.009) (0.022) (0.024) (0.001) (0.001) (0.002) (0.001) (0.001) (0.002) (0.000) (0.000)
Obs. 54,028 54,028 54,028 54,028 54,028 54,028 54,028 54,028 54,028 54,028 54,028
R-2 0.193 0.292 0.269 0.182 0.166 0.202 0.182 0.180 0.191 0.488 0.446
p-value F-tests
Post 0.26 0.01 0.00 0.37 0.93 0.63 0.07 0.00 0.00 0.51 0.94
Post× Event-Trend 0.06 0.01 0.11 0.93 0.23 0.49 0.09 0.00 0.00 0.48 0.04
46
joiner enters the VC portfolio for the first time, and indicates the first through sixth years after the joiner enters the portfolio. In Panel A, the age of the joiner is estimated as the difference between the year
the company joins the portfolio of the VC and the founding year, as reported in SDC. Young (Old) are joiners below (above) the median joiner age in the sample of two years. In Panel B, joiners are defined
as early if the first investment they secured from a VC was either at a seed or early stage round according to the SDC files, following Lindsey (2008). In Panel C, I define a joiner as new to venture capital
if the joiner is securing venture capital for the first time. In Panel D, investor experience is measured by the number of prior investments made by the VC following Lindsey (2008). Junior (Senior) are VCs
that are below (above) the median VC experience in the sample of six years. Standard errors are heteroskedasticity robust and clustered at the joiner and VC pair level.*, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
47
Table 5
Heterogeneity II: relative portfolio exchanges between joiners and portfolio companies
This table reports the coefficients and standard errors (in parentheses) from estimating Eq. (2) on the different proxies of relative portfolio exchanges. An observation is a joiner and VC pair cross time
(year). The explanatory variables of interest are Post, an indicator variable that equals one after the joiner enters the VC portfolio for the first time, and Post×Event-Trend, which equals zero before the joiner
enters the VC portfolio for the first time, and indicates the first through sixth years after the joiner enters the portfolio. In Panel A, I consider the six broad industry sectors defined by SDC: biotechnology,
communications and media, computer-related, medical, non high technology, and semiconductors (see Table 1). Standard errors are heteroskedasticity robust and clustered at the joiner and VC pair level.*,
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Dependent variable Relative
citations
received
Relative
citations
made
Relative
overall
citations
Relative
patents sold
Relative
patents
bought
Relative
patent sales
Relative
inventor
emigrates
Relative
inventor
immigrants
Relative
inventor
exchanges
Relative
alliances
Relative
mergers and
acquisitions
Panel A—Industry of joiner and portfolio companies
Same industry
Post 0.014* 0.112*** 0.126*** 0.005** 0.007** 0.012*** 0.004** 0.006** 0.009*** 0.000 -0.000
(0.008) (0.040) (0.041) (0.002) (0.003) (0.004) (0.002) (0.002) (0.003) (0.000) (0.000)
Post× Event-Trend 0.015** -0.006 0.010 0.002** 0.002 0.004* -0.000 -0.004*** -0.003*** 0.000*** 0.001***
(0.007) (0.007) (0.010) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553
R-2 0.193 0.245 0.233 0.230 0.192 0.235 0.142 0.152 0.166 0.444 0.445
Different industry
Post -0.000 -0.091*** -0.092*** -0.001 0.002*** 0.001 0.002 0.004** 0.005** -0.000 0.000
(0.003) (0.025) (0.025) (0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.000) (0.000)
Post× Event-Trend -0.001 -0.017** -0.018** 0.000 -0.000 0.000 -0.001* -0.002*** -0.003*** 0.000** 0.001***
(0.003) (0.008) (0.009) (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553
R-2 0.132 0.177 0.173 0.097 0.120 0.113 0.155 0.165 0.172 0.488 0.387
Panel B—VC fund of joiner and portfolio companies
Same fund
Post 0.008 0.007 0.015 0.005** 0.009*** 0.014*** 0.002 0.001 0.003 0.000 0.000
(0.007) (0.025) (0.026) (0.002) (0.002) (0.004) (0.002) (0.002) (0.003) (0.000) (0.000)
Post× Event-Trend 0.002 -0.001 0.001 0.002* 0.002 0.004* -0.001* -0.001 -0.002* 0.000*** 0.000*
(0.005) (0.005) (0.007) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553
R-2 0.187 0.222 0.206 0.185 0.202 0.220 0.130 0.137 0.137 0.418 0.457
Different fund
Post 0.007 0.013 0.020 -0.001 -0.000 -0.001 0.004** 0.009*** 0.011*** -0.001 -0.000
(0.005) (0.027) (0.028) (0.001) (0.002) (0.002) (0.002) (0.003) (0.003) (0.001) (0.000)
Post× Event-Trend 0.013** -0.022** -0.010 0.001 -0.000 0.001 -0.000 -0.005*** -0.004*** 0.001 0.001***
(0.005) (0.011) (0.012) (0.000) (0.001) (0.001) (0.001) (0.001) (0.001) (0.000) (0.000)
Obs. 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553
R-2 0.177 0.266 0.246 0.148 0.136 0.170 0.156 0.174 0.183 0.148 0.407
48
**, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
49
Table 6
Inventor and executive exchanges into (and from) incumbent and future portfolio companies
Panel A—Inventor exchanges
Panel B—Executive Exchanges
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Dependent variable
Relative
inventor
emigrates
Relative
Inventor
immigrants
Relative
inventor
exchanges
Relative
Inventor
emigrates
incumbent
Relative
inventor
immigrants
incumbent
Relative
inventor
exchanges
incumbent
Relative
inventor
emigrates
future
Relative
Inventor
Immigrants
future
Relative
inventor
exchanges
future
Post 0.005*** 0.008*** 0.012*** 0.001 0.007*** 0.008*** 0.003*** 0.001 0.004***
(0.002) (0.002) (0.003) (0.001) (0.002) (0.002) (0.001) (0.001) (0.001)
Post× Event-Trend -0.001* -0.006*** -0.005*** -0.001*** -0.005*** -0.005*** 0.000 -0.001** 0.000
(0.001) (0.001) (0.001) (0.000) (0.001) (0.001) (0.000) (0.000) (0.000)
Obs. 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553
R-2 0.517 0.591 0.571 0.147 0.172 0.173 0.159 0.159 0.171
Mean dep. var. 0.005 0.009 0.014 0.002 0.008 0.009 0.004 0.002 0.005
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Dependent variable Executives
emigrates
Executives
immigrants
Executive
exchanges
Executives
emigrates
incumbent
Executives
immigrants
incumbent
Executives
exchanges
incumbent
Executives
emigrates
future
Executives
immigrants
future
Executives
exchanges
future
Post 0.008*** 0.014*** 0.020*** 0.000 0.014*** 0.012*** 0.008*** 0.001 0.008***
(0.001) (0.003) (0.003) (0.001) (0.002) (0.003) (0.001) (0.001) (0.001)
Post× Event-Trend -0.002*** -0.009*** -0.010*** -0.002*** -0.008*** -0.009*** -0.000* -0.001*** -0.001***
(0.000) (0.001) (0.001) (0.000) (0.001) (0.001) (0.000) (0.000) (0.000)
Obs. 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553 121,553
R-2 0.160 0.112 0.147 0.115 0.112 0.117 0.149 0.102 0.146
Mean dep. var. 0.006 0.008 0.013 0.002 0.007 0.008 0.004 0.001 0.005
50
Panel C—Coincidence inventor and executive exchanges
This table reports the coefficients and standard errors (in parentheses) from estimating Eq. (2) on the different proxies of portfolio exchanges. An observation is a joiner and VC pair cross time (year). The
explanatory variables of interest are Post, an indicator variable that equals one after the joiner enters the VC portfolio for the first time, and Post×Event-Trend, which equals zero before the joiner enters the
VC portfolio for the first time, and indicates the first through sixth years after the joiner enters the portfolio. The dependent variable is specified at the top of each column. The sample used in the first three
columns of Panel C restricts observations to those where no executives leave the joiner to work in another portfolio company. Standard errors are heteroskedasticity robust and clustered at the joiner and VC
pair level.*, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4) (5) (6)
Dependent variable Inventor
emigrates
Inventor
immigrants
Inventor
exchanges
Conditional
inventor
emigrates
Conditional
inventor
immigrants
Conditional
inventor
exchanges
Post 0.003** 0.007*** 0.010*** 0.000 0.001*** 0.001***
(0.001) (0.002) (0.003) (0.000) (0.000) (0.000)
Post× Event-Trend -0.001 -0.005*** -0.005*** -0.000 -0.000*** -0.000
(0.000) (0.001) (0.001) (0.000) (0.000) (0.000)
Obs. 120,955 120,955 120,955 121,553 121,553 121,553
R-2 0.161 0.177 0.189 0.127 0.099 0.124
Sample Executives do
not leave
Executives do
not leave
Executives do
not leave All All All
Mean dep. var. 0.006 0.010 0.015 0.0003 0.0003 0.0006
51
Table 7
Grouped and individual inventor exchanges
Panel A—Inventor exchanges
Panel B —Circumstances surrounding emigration of teams of inventors
Panel A reports the coefficients and standard errors (in parentheses) from estimating Eq. (2) on the different proxies of portfolio exchanges. An observation is a joiner and VC pair cross time (year). The
explanatory variables of interest are Post, an indicator variable that equals one after the joiner enters the VC portfolio for the first time, and Post×Event-Trend, which equals zero before the joiner enters
the VC portfolio for the first time, and indicates the first through sixth years after the joiner enters the portfolio. The dependent variable is specified at the top of each column. Standard errors are
heteroskedasticity robust and clustered at the joiner and VC pair level.*, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. Panel B classifies the circumstances
surrounding emigration of teams of inventors from joiners to portfolio companies. There are a total of 73 Joiners for which at least one team of inventors emigrates to another portfolio company. There are
a total of 95 emigration events of teams of inventors. I classify the circumstances surrounding such migrations into seven categories of corporate events (merger or acquisition by a company outside the
VC portfolio, merger or acquisition by an incumbent portfolio company, merger or acquisition by a future portfolio company, IPO, liquidation, restructuring and founding) by comparing the timing of the
emigration and the timing of such corporate events. An emigration event is classified as a respective corporate event if it occurs within four years of the corporate event. The class “reorganization” includes
several types of corporate events including: hiring of new CEOs, securing investment from government, and opening a foreign subsidiary. An emigration event is unclassifiable if it does not take place
within four years of the six corporate events considered.
(1) (2) (3) (4) (5) (6)
Dependent variable
At least one
inventor
emigrates
At least one
inventor
immigrates
Individual
inventor
emigrates
Individual
inventor
immigrates
Team of
inventors
emigrates
Team of
inventors
immigrates
Post 0.001* 0.003*** 0.001 0.002* 0.001** 0.001**
(0.001) (0.001) (0.001) (0.001) (0.000) (0.001)
Post× Event-Trend -0.000 -0.003*** -0.000 -0.001*** -0.000 -0.001***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Obs. 121,553 121,553 121,553 121,553 121,553 121,553
R-2 0.155 0.163 0.126 0.121 0.147 0.170
Merger and acquisition outside the portfolio 21 22%
Merger and acquisition incumbent portfolio company 7 7%
Merger and acquisition future portfolio company 3 3%
Liquidation 3 3%
Foundation 34 36%
IPO 2 2%
Reorganization 7 7%
Unclassifiable 18 19%
52
Table 8
PIR Adoption and relative portfolio exchanges between joiners and portfolio companies
Panel A in this table reports the coefficients and standard errors (in parentheses) from estimating Eq. (3). An observation is a joiner and VC pair cross time, where time has been collapsed into two periods:
pre- and post-PIR adoption in the VC’s home state. The dependent variables correspond to the aggregate residuals from regressions of the outcome variables on year fixed effects, differential trends across
the home states of VCs, and fixed effects at the level of the home state of the VC, which are then collapsed to two observations per joiner and VC pair: before and after the home state of the VC adopts PIR.
PIR is an indicator that equals one after the home state of the VC adopts PIR via UPIA enactment (see Online Appendix 3). Panel B of this table reports IV estimates of Eq. (2) in the two-period panel (i.e.,
pre and post for every pair), where I instrument Post using PIR. Panel C of this table reports OLS estimates of Eq. (2) on the two-period panel. The event window includes all observations from joiner and
VC pairs for which the VC’s home state adopted PIR within the sample period and corresponds to 29 years before and 16 years after the adoption of PIR in the home state of the VC. Standard errors are
heteroskedasticity robust and clustered at the home state of the VC level throughout.*, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Dependent variable
(Residuals of):
Post
Relative
citations
received
Relative
citations
made
Relative
overall
citations
Relative
patents
sold
Relative
patents
Bought
Relative
patent
sales
Relative
inventor
emigrates
Relative
inventor
immigrants
Relative
inventor
exchanges
Relative
alliances
Relative mergers
and acquisitions
Panel A —Reduced Form
PIR 0.618*** 0.170*** 0.120*** 0.290*** 0.012*** 0.015*** 0.027*** 0.010** 0.004 0.014** 0.004*** 0.006***
(0.078) (0.052) (0.040) (0.090) (0.002) (0.003) (0.005) (0.004) (0.003) (0.006) (0.001) (0.001)
Observations 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694
R-squared 0.888 0.516 0.724 0.632 0.639 0.547 0.599 0.589 0.619 0.605 0.736 0.690
F- exc. instruments 62.76
Panel B — IV
Post 0.276*** 0.193*** 0.469*** 0.019*** 0.024*** 0.043*** 0.015*** 0.006 0.022*** 0.007*** 0.009***
(0.054) (0.044) (0.094) (0.003) (0.006) (0.009) (0.005) (0.004) (0.008) (0.001) (0.002)
Observations 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694
R-squared 0.516 0.724 0.633 0.639 0.547 0.599 0.590 0.619 0.606 0.736 0.689
Panel C —OLS
Post 0.273*** 0.196*** 0.469*** 0.018*** 0.019*** 0.037*** 0.016*** 0.007* 0.024*** 0.007*** 0.007***
(0.062) (0.054) (0.114) (0.002) (0.003) (0.004) (0.005) (0.004) (0.008) (0.001) (0.001)
Observations 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694 7,694
R-squared 0.516 0.724 0.633 0.639 0.547 0.599 0.590 0.619 0.606 0.736 0.689
53
Table 9
Robustness checks: PIR adoption and relative portfolio exchanges between joiners and portfolio companies
(1) (2) (3) (4) (5) (6) (7) (8) (9)/ (10) (11) (12)
Dependent variable
(residuals of):
Post
Relative
citations
received
Relative
citations
made
Relative
overall
citations
Relative
patents
sold
Relative
patents
bought
Relative
patent
sales
Relative
inventor
emigrates
Relative
inventor
immigrants
Relative
inventor
exchanges
Relative
alliances
Relative mergers
and acquisitions
Panel A—Joiners and portfolio companies out of the home state of the VC
Reduced form
PIR 0.578*** 0.077** 0.055** 0.132*** 0.011*** 0.013*** 0.024*** 0.001 0.001 -0.002 0.002* 0.004***
(0.056) (0.034) (0.025) (0.047) (0.002) (0.004) (0.006) (0.003) (0.003) (0.003) (0.001) (0.001)
Observations 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870
R-squared 0.878 0.521 0.664 0.611 0.720 0.556 0.622 0.631 0.527 0.641 0.652 0.791
F- exc. instruments 106.53
OLS
Post 0.089 0.073 0.162* 0.018*** 0.017*** 0.035*** 0.004 0.003 -0.002 0.003** 0.004***
(0.055) (0.044) (0.083) (0.004) (0.004) (0.007) (0.005) (0.003) (0.004) (0.001) (0.001)
Observations 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870
R-squared 0.520 0.663 0.610 0.720 0.555 0.622 0.631 0.527 0.641 0.652 0.790
IV
Post 0.133* 0.096* 0.228** 0.018*** 0.023*** 0.041*** 0.001 0.001 -0.004 0.003** 0.007***
(0.067) (0.048) (0.098) (0.003) (0.007) (0.010) (0.006) (0.004) (0.004) (0.002) (0.002)
Observations 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870 3,870
R-squared 0.520 0.663 0.610 0.720 0.555 0.622 0.631 0.527 0.641 0.652 0.790
Panel B— Joiners and portfolio companies out of the home state of the VC and out of coincidental states
Reduced form
PIR 0.568*** 0.068** 0.052** 0.120*** 0.012*** 0.011*** 0.023*** 0.000 0.000 -0.003 0.001 0.004***
(0.054) (0.025) (0.024) (0.040) (0.004) (0.003) (0.006) (0.004) (0.003) (0.003) (0.001) (0.001)
Observations 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618
R-squared 0.879 0.681 0.524 0.625 0.566 0.722 0.632 0.634 0.528 0.645 0.654 0.801
F- exc. instruments 110.64
OLS
Post 0.097** 0.070* 0.167** 0.015*** 0.018*** 0.034*** 0.003 0.002 -0.003 0.002 0.004***
(0.043) (0.037) (0.070) (0.004) (0.004) (0.007) (0.006) (0.004) (0.004) (0.002) (0.001)
Observations 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618
R-squared 0.681 0.524 0.624 0.565 0.722 0.632 0.635 0.528 0.645 0.654 0.800
IV
Post 0.120** 0.092* 0.212** 0.022*** 0.019*** 0.041*** 0.000 0.001 -0.005 0.003 0.007**
(0.051) (0.047) (0.084) (0.008) (0.004) (0.011) (0.007) (0.005) (0.005) (0.002) (0.003)
Observations 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618 3,618
R-squared 0.681 0.524 0.624 0.565 0.722 0.632 0.634 0.528 0.645 0.654 0.800
54
The first block of each panel in this table reports the coefficients and standard errors (in parentheses) from estimating Eq. (3). An observation is a joiner and VC pair cross time, where time has been
collapsed into two periods: pre- and post-PIR adoption in the VC’s home state. The dependent variables correspond to the aggregate residuals from regressions of the outcome variables on year fixed
effects, differential trends across the home states of VCs, and fixed effects at the level of the home state of the VC, which are then collapsed to two observations per joiner and VC pair: before and after
the home state of the VC adopts PIR. PIR is an indicator that equals one after the home state of the VC adopts PIR via UPIA enactment. The second block of each panel in this table reports OLS estimates
of Eq. (2) on the two-period panel (i.e., pre and post for every pair). The third block of each panel in this table reports IV estimates of Eq. (2) in the two-period panel, where I instrument Post using PIR.
The event window includes all observations from joiner and VC pairs for which the VC’s home state adopted PIR within the sample period, and corresponds to 29 years before and 16 years after the
adoption of PIR in the home state of the VC. Panel A restricts the sample to joiners and portfolio companies that are located out of the home state of the VC. Panel B restricts the data to joiners and portfolio
companies that are located out of the home state of the VC and are also not located in states that coincidentally passed PIR at the same time as the home state of the VC (coincidental states). Standard errors
are heteroskedasticity robust and clustered at the home state of the VC level throughout.*, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
Top Related